How do I leverage Artificial Intelligence (AI)?

From PRIMO.ai
Jump to navigation Jump to search

YouTube ... Quora ...Google search ...Google News ...Bing News


The future is already here. It’s just not very evenly distributed. - William Gibson


"Leveraging AI" means using artificial intelligence to get more value out of your time, skills, data, or business than you could without it. In 2026 that covers three overlapping routes. You can use AI tools to do your current work better or faster. You can build AI into products, services, and workflows for other people. And you can sell AI-related expertise: consulting, training, implementation, data, or content. This page collects courses, monetization models, use-case patterns, roadmaps, and talks for all three, and adds the current research on what actually pays off.

The question matters because general-purpose AI has spread faster than almost any earlier technology. Stanford's 2026 AI Index reports that generative AI reached about 53 percent population adoption within three years, faster than the personal computer or the internet, and that 88 percent of organizations now use AI in at least one business function. Yet the same report, along with McKinsey's 2025 global survey, finds that only a small minority of organizations have scaled AI to the point of measurable profit impact. Access is no longer the bottleneck. Knowing where AI helps, where it fails, and how to reorganize work around it is.

The people and organizations shaping this field fall into a few groups. Model and platform providers such as OpenAI, Anthropic, Google, Microsoft, Meta, DeepSeek, and Hugging Face supply the underlying capabilities. Educators such as Andrew Ng (DeepLearning.AI and AI Fund) and Andrej Karpathy translate those capabilities into practical skills. Economists and management researchers measure what AI does to productivity, wages, and jobs: Erik Brynjolfsson and colleagues at Stanford's Digital Economy Lab, Fabrizio Dell'Acqua and colleagues at Harvard Business School working with Ethan Mollick of Wharton, and Anders Humlum at the University of Chicago, among others.

The central mechanism is simple. AI sharply lowers the cost of certain cognitive tasks: drafting, summarizing, classifying, translating, searching, and writing routine code. When something gets cheap, value shifts to whatever remains scarce: domain expertise, judgment about what is correct, proprietary data, trust, distribution, and the ability to redesign a workflow so the cheap part actually feeds something valuable.

That is the paradigm shift from the earlier machine learning era. In the 2010s, leveraging AI usually meant hiring data scientists to collect labeled data and spend months training a custom model. Today most value comes from adapting foundation models through prompting, retrieval, tool connections, and agents, often in days. The competitive edge has moved from model building to integration, evaluation, and change management. Pricing is shifting too, from per-seat software licenses toward charging per task or per outcome. See Case Studies for applied examples.


AI makes some kinds of work dramatically cheaper, but cheap output is not the same as value. The people and businesses that benefit most pick the right tasks, verify the results, redesign the work around them, and own something AI cannot easily copy.


The Mechanism

Leveraging AI follows a repeatable sequence, whether you are an employee, a freelancer, or a company.

  1. Break work into tasks. Jobs are bundles of tasks, and AI affects tasks, not whole jobs. List what you or your client actually do each week and roughly how long each task takes.
  2. Locate the frontier. AI capability is uneven. Researchers call this the "jagged frontier": models excel at some tasks that look hard and fail at others that look easy. Test candidate tasks directly instead of assuming.
  3. Choose a mode: augmentation or automation. In augmentation, a person works with AI iteratively: brainstorming, drafting, getting feedback, learning. In automation, the person delegates a task and accepts the output, possibly after review. Anthropic's Economic Index classifies real usage this way and finds the two roughly balanced; in its January 2026 report, 52 percent of Claude.ai conversations were augmentative and 45 percent automated.
  4. Give the model context and tools. Most business value requires connecting a model to your own information and systems: documents through Retrieval-Augmented Generation (RAG), instructions through Prompt Engineering (PE), specialized behavior through Fine-tuning, and actions through tool connections and agents. The Model Context Protocol (MCP) has become the common standard for these connections.
  5. Evaluate and review. Define what "good" looks like, test on real examples, and decide where a human checks the output. This step is what separates working deployments from abandoned pilots.
  6. Redesign the workflow. Time saved only becomes value if it is reinvested: more output, faster turnaround, higher quality, or new services. In McKinsey's 2025 survey, the small group of AI high performers was nearly three times as likely as others to have fundamentally redesigned workflows.
  7. Capture the value. For an employee, that may mean a promotion, a new role, or a stronger negotiating position. For a freelancer or business, it means pricing: hourly, per project, subscription, usage-based, or per outcome.

Dependencies and conditions. Results depend on data quality and access, integration with existing systems, the reliability the task demands (a draft email tolerates errors; a tax filing does not), user skill, organizational support, and the legal environment.

Where uncertainty remains. Measurement is hard. People routinely overestimate how much AI helps them, as METR's 2025 developer study showed, and many organizations do not track outcomes at all. Capabilities also change every few months, so last quarter's limitation may already be gone, and last quarter's competitive advantage may already be commoditized.

The Physics

For a practical topic like this, the "physics" is the set of technical and economic constraints that behave like laws. They determine what works regardless of how a product is marketed.

Models are probabilistic. A Large Language Model (LLM) generates text one token at a time by predicting likely continuations, conditioned on everything in its context window. Outputs vary between runs and can be confidently wrong. That makes verification part of every serious use case, not an optional extra.

Reliability compounds. When an agent chains steps, errors multiply. If each step succeeds with probability p, a task of n independent steps succeeds with probability

<math>P_{\text{success}} = p^{n}</math>

At 95 percent per step, a ten-step task succeeds only about 60 percent of the time; at 99 percent per step, about 90 percent. That is why agentic workflows need checkpoints, retries, and human review at the riskiest steps, and why improving per-step reliability matters more than adding steps.

The cost of intelligence keeps falling. Stanford's 2025 AI Index estimated that the price of querying a model with GPT-3.5-level performance fell from about $20 to about $0.07 per million tokens between November 2022 and October 2024, a drop of more than 280 times. Cheaper inference makes new products viable, but it also commoditizes anything that is only a thin layer over a model. If your product's main value is access to a capability, the next price cut or platform feature can erase it.

Reasoning and agents cost more per task. Newer reasoning models generate long internal chains of thought, and agents call models many times per task. Price per token falls while tokens per task rise, so the total cost per completed task has to be measured, not assumed.

The economics of a use case. A simple way to decide whether AI helps with a task:

<math>\text{Net value} = T_{\text{saved}} \times V_{\text{time}} - C_{\text{review}} - p_{\text{error}} \times C_{\text{error}} - C_{\text{tools}}</math>

Here Tsaved is the time saved, Vtime is what that time is worth, Creview is the cost of checking the output, perror × Cerror is the expected cost of mistakes that slip through, and Ctools covers subscriptions, API fees, and integration. Many failed projects saved time on paper but ignored the review and error terms.

Commoditization and complements. When AI makes a kind of output abundant, its market price tends to fall, and value migrates to complements that stay scarce: trusted brands, proprietary data, distribution, regulatory know-how, physical-world execution, and accountable human judgment. That helps explain why generic AI writing services face falling prices while domain experts who evaluate AI output command premium rates.

Gains do not automatically reach paychecks. Humlum and Vestergaard's Danish data found that chatbot users reported average time savings of about 2.8 percent of work hours, and that only a small share of productivity gains passed through to earnings. Saved time becomes income only if you, or your employer, redeploy it to work someone pays for.

Why This Matters

  • For individuals. PwC's 2025 Global AI Jobs Barometer, built on close to a billion job ads, found that roles requiring AI skills carried an average 56 percent wage premium over similar roles without them, up from 25 percent the year before, and that the skills employers seek in AI-exposed jobs are changing 66 percent faster than in other jobs. At the same time, Stanford research shows early-career workers in the most AI-exposed occupations losing ground. Using AI well is becoming a baseline expectation rather than a differentiator.
  • For businesses. Adoption is nearly universal, but the gap between experimenting and capturing value is wide. The organizations that pull ahead invest in workflow redesign, visible leadership ownership, and measurement, not just licenses.
  • For entrepreneurs and freelancers. Falling costs open niches that were uneconomic a few years ago, but they also flood markets with low-cost competitors. Durable businesses combine AI with something hard to replicate.
  • For consumers. The "make money with AI" boom has attracted outright fraud, and regulators and platforms are responding with enforcement actions, labeling rules, and demonetization of mass-produced content.
  • Limitations. The aggregate economic effects measured so far are modest. Large productivity gains appear in specific tasks and settings, not uniformly across the economy.
  • What would change if the optimists are right. If agents become reliable enough for long, multi-step work, the main bottleneck shifts from doing tasks to specifying, supervising, and taking responsibility for them, and the premium on judgment and domain expertise rises further.
  • What remains uncertain. Whether entry-level hiring declines persist, how quickly reliability improves, how regulation evolves, and whether productivity gains show up in national statistics. See Risk, Compliance and Regulation and Help Wanted.

Recent Developments

Adoption is broad; scaled value is rare

McKinsey's State of AI 2025. McKinsey surveyed 1,993 participants in 105 countries between late June and late July 2025. Eighty-eight percent reported regular AI use in at least one business function, up from 78 percent a year earlier. But roughly two-thirds said their organizations had not yet begun scaling AI across the enterprise; only about 7 percent described AI as fully scaled. About 6 percent qualified as "high performers," attributing at least 5 percent of EBIT to AI and reporting significant value. Those high performers were roughly three times as likely to pursue transformative change, to redesign workflows fundamentally, and to report strong senior-leader ownership, and more than a third of them spent over 20 percent of their digital budgets on AI. On agents, 62 percent of respondents were at least experimenting and 23 percent were scaling an agentic system somewhere, but in any single business function no more than 10 percent were scaling agents. The survey is self-reported and skews toward larger organizations, but its direction matches other evidence.

Stanford AI Index 2026. The April 2026 edition reported that generative AI reached about 53 percent population adoption within three years and that organizational adoption rose to 88 percent. It estimated that generative AI delivered about $172 billion a year in value to US consumers by early 2026. It also documented 362 AI incidents in 2025, up from 233 the year before, and noted that full integration at scale remains rare across business functions.

The "95 percent" debate. MIT Project NANDA's report The GenAI Divide: State of AI in Business 2025 (summer 2025) claimed that 95 percent of organizations it studied saw no measurable profit-and-loss return from generative AI, while about 5 percent of integrated pilots produced millions in value. The research drew on 52 interviews, 153 survey responses gathered at industry conferences, and a review of more than 300 public AI initiatives from January through June 2025. Critics noted that "success" meant marked P&L impact within roughly six months, that individual productivity gains from tools like ChatGPT were excluded, that the underlying data was not released, and that the authors' own agent-infrastructure work shaped their recommendations. A 2025 Wharton survey of enterprise leaders reported a very different picture, with roughly three-quarters reporting positive ROI. What the NANDA report does establish: custom enterprise pilots frequently stall, usually because tools do not learn, integrate, or fit workflows. What it does not establish: that AI "doesn't work."

What the productivity evidence actually shows

  • Customer support (field study). Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied the staggered rollout of a generative AI assistant to 5,179 support agents at a large software firm, published in the Quarterly Journal of Economics in 2025. Issues resolved per hour rose about 14 percent on average and about 34 percent for novice and lower-skilled agents, with little effect on the most experienced. Customer sentiment and employee retention improved. The AI appeared to spread the practices of top performers to newer workers. Limits: one firm, one language, text chat only.
  • Consulting tasks (field experiment). Fabrizio Dell'Acqua and colleagues assigned 758 Boston Consulting Group consultants to work with or without GPT-4. On tasks inside the AI's capability frontier, consultants with AI completed about 12 percent more tasks, about 25 percent faster, with more than 40 percent higher quality ratings. On a task deliberately chosen to fall outside the frontier, consultants using AI were about 19 percentage points less likely to reach the correct answer. The lesson: knowing where the frontier lies is itself a core skill.
  • Software development (randomized trials). In July 2025, METR reported a randomized trial with 16 experienced open-source developers completing 246 real tasks in large repositories they had worked on for years. Before starting, they forecast that AI would make them 24 percent faster; with early-2025 tools, they were measured as 19 percent slower, and still believed afterward that AI had helped. Contributing factors included time spent prompting, waiting, and reviewing suggestions they often rejected. METR cautioned that results apply most strongly to experts on familiar, mature codebases. Its early-2026 follow-up found some evidence of speedups with newer tools, but recruitment problems (many developers refused to work without AI) made the estimate unreliable. A separate randomized trial with 96 Google engineers found AI shortened time on an enterprise task by about 21 percent, with wide confidence intervals.
  • The whole labor market (administrative data). Anders Humlum and Emilie Vestergaard linked large adoption surveys covering about 25,000 workers and 7,000 workplaces in 11 AI-exposed occupations in Denmark to official earnings and hours records. Despite widespread, employer-encouraged chatbot use, they found precise null effects on earnings and recorded hours, ruling out average effects larger than about 1 to 2 percent two years after adoption. A 2026 revision, retitled Still Waters, Rapid Currents, emphasizes that adoption is already reshaping tasks and prompting occupational switching before it shows up in pay.
  • Real-world usage patterns. Anthropic's Economic Index, built from privacy-preserving analysis of millions of Claude conversations, found in its January 2026 report (covering November 2025) that the ten most common tasks made up 24 percent of usage, that augmentation was again slightly more common than automation, and that per-capita usage across countries remains strongly correlated with income.

What this establishes: gains are real but task-specific; they are largest for less experienced workers and well-defined tasks, smaller or even negative for experts doing complex work they already know well. What it does not establish: a single productivity multiplier you can apply to your job. Self-reported gains are unreliable, so measure.

The labor market: early-career and freelance effects

Entry-level workers. Brynjolfsson, Bharat Chandar, and Ruyu Chen's Canaries in the Coal Mine? uses payroll records from ADP, the largest US payroll provider. The August 2025 version found that workers aged 22 to 25 in the most AI-exposed occupations, such as software development and customer service, saw a 13 percent relative employment decline since late 2022, after controlling for firm-level shocks, while older workers in the same occupations and young workers in less-exposed fields held steady or grew. Declines were concentrated where AI automates rather than augments work, and showed up in headcount rather than pay. Later revisions widened the gap: the authors reported in August 2026 that the relative decline had reached 19 percent by June 2026, while still finding no widespread job displacement overall. The evidence is correlational, and other factors could contribute, though the authors' robustness checks (excluding tech firms and remote-friendly occupations) did not change the pattern.

Freelancers. Xiang Hui, Oren Reshef, and Luofeng Zhou, publishing in Organization Science, found that after ChatGPT's release, writing-related freelancers on Upwork saw monthly jobs fall about 2 percent and monthly earnings fall about 5.2 percent, with similar effects for image work after image generators launched. Highly rated freelancers were not spared. Research by Ozge Demirci and colleagues found postings in automation-prone freelance categories fell roughly 21 percent. At the same time, Upwork reported fast-rising demand for AI-related freelance skills, including AI integration, AI video work, and data annotation. Commodity freelance work is shrinking; work that applies or supervises AI is growing.

Agents, MCP, and new building blocks

Anthropic introduced the Model Context Protocol in late 2024 as an open standard for connecting AI applications to data sources and tools. By December 2025 the project reported more than 97 million monthly SDK downloads, about 10,000 active servers, and client support across ChatGPT, Claude, Cursor, Gemini, Microsoft Copilot, and Visual Studio Code. On December 9, 2025, Anthropic donated MCP to the new Agentic AI Foundation, a directed fund of the Linux Foundation co-founded by Anthropic, Block, and OpenAI with support from Google, Microsoft, AWS, Cloudflare, and Bloomberg; Block's goose and OpenAI's AGENTS.md joined as founding projects. For anyone leveraging AI, this matters because one integration now works across many AI products, and building MCP servers for business systems has become a distinct service category.

In June 2025, Andrej Karpathy described this era as "Software 3.0," in which natural language becomes a programming interface, and he had coined "vibe coding" months earlier for building software largely by describing it to AI. Andrew Ng's talk at the same event argued that AI coding tools have made prototyping so fast that the bottleneck shifts to product decisions and user feedback. Both talks are listed under Additional Viewing.

New business models and pricing

  • Outcome-based pricing. Intercom's Fin customer service agent charges about $0.99 per resolved conversation instead of a per-seat fee, and industry reports in 2026 described Fin approaching $100 million in annual recurring revenue. Analysts compare that to human-handled tickets that often cost $5 to $15. The catch: how a "resolution" is defined drives the bill, and some customers report unpredictable costs. If you sell AI services, pricing by outcome can win clients, but only if you can measure outcomes credibly.
  • Expert data work. AI labs now pay heavily for domain experts to create training data and evaluate model outputs. Meta invested about $14.3 billion for a 49 percent stake in Scale AI in June 2025, and in October 2025 Mercor raised $350 million at a $10 billion valuation, reporting more than 30,000 contractors averaging over $85 per hour. For doctors, lawyers, accountants, engineers, and other specialists, this is a real income stream. Read intellectual-property terms carefully, and recognize that demand depends on labs' budgets and that the work may help automate parts of your own field.
  • Thin wrappers face pressure. As inference costs fall and major platforms add features, products whose only value is a nicer interface over someone else's model get squeezed. Durable AI businesses own data, distribution, workflow integration, or accountability that customers cannot get directly from the model provider.

Guardrails: scams, platform rules, and regulation

  • Fraud enforcement. In September 2024 the US Federal Trade Commission launched Operation AI Comply, announcing five actions and stating that there is no AI exemption from existing law. Several targeted "AI-powered" business opportunity schemes: Ascend Ecom, which allegedly promised passive income from AI-run online stores; Ecommerce Empire Builders, which allegedly sold training costing nearly $2,000 and "done for you" stores costing tens of thousands of dollars; and FBA Machine (also called Passive Scaling), which allegedly promised guaranteed income. The sweep also covered a "robot lawyer" service and a tool used to generate fake reviews.
  • Platform rules. On July 15, 2025, YouTube clarified its Partner Program monetization policy on "inauthentic" content to better identify mass-produced and repetitive videos. The update does not ban AI tools, but channels built on templated AI voiceovers and recycled visuals with little human contribution risk losing monetization. Realistic synthetic media also requires disclosure.
  • EU AI Act. The Act's AI literacy duty for organizations deploying AI has applied since February 2025, and obligations for general-purpose model providers since August 2025. The transparency obligations of Article 50 became enforceable on August 2, 2026: users must be told when they are interacting with an AI system, providers of generative AI must mark synthetic output in machine-readable form, and deployers must label deepfakes and AI-generated text on matters of public interest unless a person takes editorial responsibility. The Digital Omnibus on AI, in force since July 2026, postponed the high-risk system requirements to December 2027 and gave systems already on the market until December 2, 2026 to implement machine-readable marking, but left the other Article 50 duties in place. The rules apply to non-EU providers and deployers whose output is used in the EU, which includes many freelancers and content businesses.

Open questions

  • Will the entry-level hiring declines seen in AI-exposed occupations persist, spread, or reverse as firms adapt?
  • How quickly will agents become reliable enough for long, unsupervised tasks, and which tasks will remain human-supervised?
  • Why do self-reported productivity gains so often exceed measured ones, and which measurement methods should businesses trust?
  • Which AI businesses have durable advantages once model access is cheap and universal?
  • Will productivity gains reach wages broadly, or concentrate among firms and workers who already have complementary assets?
  • How will labeling rules and platform policies reshape the economics of AI-generated content?

Courses & Certifications

To effectively leverage AI, it is critical to align your learning path with your professional goals. We have categorized the following resources to help you navigate your journey.

Two practical notes for 2026. First, high-quality learning material is now abundant and often free, so the scarce thing is applied practice: a portfolio of real projects, measured results, and write-ups of what failed usually carries more weight with employers and clients than a stack of certificates. Second, certifications are most useful as structured study plans and as signals in fields where buyers cannot easily judge skills directly, such as cloud platforms and regulated industries.

Beginner: Prompt Engineering

  • Prompt Engineering for ChatGPT - Practical techniques for optimizing model outputs.
  • AI for Everyone | DeepLearning.AI - Andrew Ng's non-technical introduction to what AI can and cannot do, how to spot opportunities, and how to plan AI projects.
  • Google Cloud Generative AI Leader certification - Launched in May 2025 for managers and other non-technical professionals; a $99, 90-minute exam with a free prep path of roughly seven to eight hours covering generative AI concepts, business use cases, and agents.
  • AWS Certified AI Practitioner - Foundational certification covering AI, machine learning, and generative AI concepts and responsible use on AWS.
  • Andrej Karpathy's How I Use LLMs (see Additional Viewing) - A practical tour of chat assistants, reasoning models, search, file uploads, coding tools, and voice, aimed at a general audience.

Intermediate: AI Development

Advanced: MLOps

Choosing a path

Your goal Start with Prove it with
Use AI better in your current job AI for Everyone, Prompt Engineering for ChatGPT, Karpathy's How I Use LLMs A before-and-after measurement of time or quality on three real tasks
Lead AI adoption on a team Google Generative AI Leader or AWS AI Practitioner, plus the McKinsey and Stanford reports in References A pilot with a defined success metric, review process, and decision to scale or stop
Build AI applications Hugging Face LLM, Agents, and MCP courses; Python fundamentals A deployed app with retrieval, tool use, and an evaluation set you can show
Work on models and infrastructure Machine Learning Specialization, Zero to Hero, Machine Learning in Production Reproduced results, Kaggle work, or contributions to open-source projects
Sell AI services Any of the above, plus the monetization section below Case studies with client-verified outcomes

How to Monetize AI and Machine Learning Solutions

What has changed. Most of the ideas below still work, but the ground has shifted since generative AI went mainstream. Model access is cheap and universal, so simply "offering AI" is no longer a business. Clients increasingly want measured outcomes rather than demos. Agents and standards like MCP have created new integration work. Domain experts can earn directly by training and evaluating models. And regulators and platforms now police deceptive "AI income" claims and mass-produced content. The strongest opportunities pair AI with something you already own: expertise, relationships, data, or a distribution channel.

There are several ways you can make money in AI:

  • AI Consulting and Integration Services: Offer consulting services to help businesses understand how AI can benefit their operations. You can offer consulting services to businesses that want to leverage AI but don't have the expertise or resources to do so. As an AI consultant, you can help businesses identify opportunities for AI, develop AI strategies, and implement AI solutions. Provide guidance on AI strategy, implementation, and integration into existing systems. The research above suggests the most valuable consulting is not tool selection but workflow redesign, evaluation design, and change management, which is where most stalled pilots fail.
  • Training: You can offer training services to individuals or businesses that want to learn more about AI. This could include online courses, workshops, or one-on-one training sessions. Demand is reinforced by regulation: the EU AI Act requires organizations deploying AI to ensure a sufficient level of AI literacy among staff, which has applied since February 2025.
  • Freelancing: You can offer your AI skills as a freelancer on platforms like Upwork, Freelancer, or Fiverr. There are many businesses and individuals looking for AI expertise on these platforms, and you can offer your services for a fee. Expect falling prices for commodity writing, design, and simple coding gigs, which research links to generative AI, and rising demand for AI integration, automation, and data annotation work.
  • Data Monetization: If you have access to valuable and well-curated data, you can offer data-driven insights, trends, or predictions to businesses willing to pay for this information. You can offer data labeling services to businesses that need high-quality labeled data for their AI models. This involves manually labeling data to improve the accuracy of machine learning models. The market has moved upmarket: AI labs now pay domain experts, often well above typical labeling rates, to write training examples and grade model outputs, through firms such as Mercor, Scale AI, and Surge AI.
  • Personalization and Recommendations: Use AI algorithms to deliver personalized content, recommendations, or targeted advertising, which can attract businesses looking to enhance user engagement and increase conversions.
  • Process Automation: Implement AI-driven automation solutions to streamline and optimize business processes for other companies. This could include automating customer support, data entry, or supply chain management, leading to increased efficiency and cost savings. Combining language models with traditional robotic process automation lets automations handle unstructured inputs such as emails and documents.
  • AI Agent Development and Integration: Build agents that complete multi-step tasks inside a client's systems, and write MCP servers that connect models to business data and tools. Because errors compound across steps, the valuable skill is designing checkpoints, evaluation, and human review, not just wiring up a model.
  • AI Evaluation, Governance, and Compliance: Help organizations test AI systems for accuracy, bias, security weaknesses such as prompt injection, and regulatory requirements, including EU AI Act transparency and documentation duties. See Risk, Compliance and Regulation.
  • Partnerships and Collaborations: Collaborate with existing companies to integrate AI capabilities into their products or co-create AI-driven solutions for mutual benefit.
  • AI in IoT (Internet of Things): Combine AI with IoT devices to create smart and autonomous systems for home automation, industrial applications, or healthcare monitoring.
  • AI-driven Products or Services: Develop AI-powered products or services that address specific needs or pain points in the market. These could be AI-based software applications, chatbots, virtual Assistants, recommendation systems, or predictive analytics tools. You can create AI-powered products or tools that solve specific business problems. For example, you could create a chatbot that automates customer support or an analytics tool that uses machine learning to identify trends and patterns in data. Avoid building a thin wrapper around a general model; the next platform feature or price cut can erase it.
    • AI as a Service (AIaaS): Offer AI capabilities as a service to other businesses. This could involve providing access to pre-trained machine learning models, natural language processing (NLP) APIs, or computer vision algorithms that they can integrate into their own products or workflows.
    • Subscription-based AI Services: Offer subscription plans for access to premium AI features, datasets, or AI-powered tools on a recurring basis.
    • Usage-based and Outcome-based Pricing: Charge per task completed or per result delivered, as Intercom does with roughly $0.99 per resolved support conversation. This aligns price with value but requires a clear, auditable definition of success.
  • Market Research: Provide AI-powered market research and consumer behavior analysis for businesses seeking valuable insights.
  • Social Media: Use AI to analyze social media data for sentiment analysis, trend identification, or targeted advertising.
    • Content Creation: Utilize AI to generate content, such as automated article writing, video editing, or music composition. Platforms now penalize mass-produced, repetitive content (YouTube tightened its monetization policy in July 2025), and the EU requires labeling of deepfakes and of AI-generated text on matters of public interest. Sustainable content businesses use AI to amplify a distinct human voice, not replace it.
      • Virtual Reality (VR) and Augmented Reality (AR): Combine AI with VR and AR technologies to create immersive and intelligent virtual experiences.
      • Music and Entertainment: Use AI to compose music, create personalized playlists, or enhance audio and visual effects in entertainment content. Check the licensing terms of the tools you use and the rules of each distribution platform before selling AI-generated works.
  • - Specific industries, AI in... -
    • Sports Analytics: Provide AI-driven sports analytics services to sports teams and organizations for performance optimization, player scouting, and fan engagement.
    • Healthcare: Develop AI-powered healthcare solutions, such as medical image analysis, disease diagnosis, drug discovery, or personalized treatment plans.
      • Healthcare Management: Create AI applications for hospital resource management, patient flow optimization, or medical appointment scheduling.
      • Personalized Medicine: Offer AI-driven solutions for precision medicine, including genetic analysis and personalized treatment recommendations.
    • E-commerce: Implement AI-driven product recommendations, personalized shopping experiences, or AI-powered customer service for e-commerce platforms.
    • Finance: Create AI-driven financial tools, like robo-advisors for investment management, fraud detection systems, or credit risk assessment models.
    • Personal Finance: Offer AI-powered budgeting tools, financial planning, or expense tracking applications for individuals.
    • Gaming and Entertainment: Build AI-powered features for gaming, such as intelligent NPCs (non-playable characters), dynamic storylines, or procedural content generation.
    • Education: Develop AI-based educational platforms that offer personalized learning paths, automated grading, or intelligent tutoring systems.
    • Marketing and Advertising: Utilize AI to enhance marketing campaigns through better audience targeting, content optimization, or social media analytics.
    • Cybersecurity: Develop AI-powered cybersecurity solutions to detect and prevent cyber threats, malware, and data breaches.
    • Agriculture: Create AI applications for precision agriculture, including crop monitoring, yield prediction, and automated farming equipment.
    • Transportation: Build AI systems for autonomous vehicles, traffic optimization, route planning, or predictive maintenance for transportation companies.
    • Supply Chain and Logistics: Develop AI-based solutions for inventory management, demand forecasting, route optimization, or supply chain risk analysis.
    • Real Estate: Use AI to analyze property data and provide insights for real estate investments, property valuation, or personalized property recommendations.
    • Human Resources: Offer AI-based recruitment and talent management tools, employee sentiment analysis, or automated HR processes.
    • Energy Management: Develop AI solutions to optimize energy consumption, predictive maintenance for energy infrastructure, or demand forecasting for energy providers.
    • Environmental Monitoring: Develop AI solutions for environmental data analysis, climate modeling, or wildlife tracking and conservation.
    • Government and Public Services: Develop AI applications to improve public services, such as traffic management, waste management, or emergency response systems.

Load video
YouTube


Matching a business model to a moat

Business model What customers pay for What protects it Main risk
Consulting and integration Getting AI to work inside their workflows Domain knowledge, client relationships, evaluation skill Projects stall without workflow redesign
Training Faster, safer adoption by staff Teaching quality, sector specialization, compliance know-how Free courses compete on price
Freelancing Completed deliverables Specialization and reputation Commodity gigs face falling prices
Expert data work Labeled data and model evaluations Rare domain expertise Demand tied to AI lab budgets; IP terms
AI products and SaaS Solving a specific problem end to end Proprietary data, integrations, distribution Thin wrappers erased by platform features
Outcome-priced agents Resolved tasks Measured reliability and workflow ownership Disputes over what counts as an outcome
Content and media Attention and audience trust Distinct voice and community Platform demonetization and labeling rules

Red flags: avoiding "make money with AI" schemes

The FTC's 2024 Operation AI Comply cases show a recurring pattern. Be wary of any offer that:

  • Promises passive or guaranteed income from "AI-powered" stores, trading bots, or content channels.
  • Charges thousands of dollars for "done for you" businesses or coaching before you have seen verifiable results from real customers.
  • Relies on screenshots of earnings, countdown timers, and testimonials you cannot independently check.
  • Claims an AI tool replaces licensed professionals such as lawyers or financial advisers.
  • Encourages fake reviews or undisclosed AI-generated endorsements, which the FTC treats as deceptive.

AI will be the greatest wealth creator in history. ... It’s going to destroy barriers
- Matt Higgins, a self-made millionaire, CEO of investment firm RSE Ventures and guest star on ABC’s Shark Tank

That optimism may prove right at the scale of the whole economy, but the evidence so far shows gains concentrating among people and firms that combine AI with complementary skills, data, and workflow changes, while commodity work faces falling prices. Treat wealth-creation claims as a reason to learn, not a guarantee of income.

Load video
YouTube
Lessons from X.ai Founder Dennis Mortensen - Building An Artificial Intelligence Startup
X.ai is an NYC based artificial intelligence company that’s raised over $30 million. I sat down with the founder Dennis Mortensen to learn more about how they’re positioning and selling the product to the enterprise. Talking to him got me super curious about the rise of artificial intelligence, and I hope this interview does similar for you.

Load video
YouTube
The Power of Machine Learning and The Opportunity for Monetization
Learn more about AWS for Media & Entertainment at: https://amzn.to/2PL2QZp From the 2019 NYC M&E Symposium: As interest in AI and ML continues to grow, we have seen many companies investigate the use of machine learning to help transform their content operations. By leveraging the power of this technology, Videofashion is able to find targeted content within 18,000 hours of fashion-focused television programming and videos using Curio™, by GrayMeta. With this rich library, Videofashion has teamed up with GrayMeta to provide an innovative platform which delivers deep insights into all their content, making it accessible and searchable online. Join GrayMeta and Videofashion along with AWS to further discuss machine learning and the monetization of assets.

Load video
YouTube
The Only 14 Ways to Make Money with AI in 2026
A breakdown of high-potential business models leveraging the latest generative AI and agentic workflows to solve real-world problems.

Load video
YouTube
7 Ways to make money using Machine Learning and AI | Make MONEY from Machine Learning
This video titled "7 Ways to make money using Machine Learning and AI | Make MONEY from Machine Learning" explains 7 ways to make money using Machine Learning and AI (artificial intelligence). Opportunities are endless and there are a lot of ways using which people like students, professionals, teachers who has acquired knowledge in this field can earn a lot of money.

Example Patterns to Leverage

There are many ways to leverage AI to improve your work, business, and life. Here are a few common patterns/categories where any AI Use Cases falls irrespective of industries.:

  • Generative AI/Conversational AI: When a machine establishes a conversation with human in natural way. Example, ChatGPT, Claude, Gemini, Siri, Alexa, Cortana. Natural Language Processing (NLP) is a branch of AI that deals with the interaction between computers and human language. By leveraging NLP, you can use AI-powered chatbots to interact with customers and provide support or use Sentiment Analysis to gauge customer feedback.
  • Assistants, Personal Companions & Agents: Enhance our lives through intelligent support, personalized assistance, and efficient task management.
  • Agentic Workflows: Systems that plan and execute multi-step tasks using tools, such as researching a topic, filling forms, reconciling records, or updating a CRM. Connect them to business systems through MCP, add checkpoints where errors would be costly, and measure end-to-end success rates rather than demo quality.
  • AI-Assisted Software Development: AI pair programming tools and coding agents write, explain, test, and refactor code, and let non-programmers build simple tools through codeless and natural-language interfaces. Evidence on speed is mixed and depends on the developer, codebase, and tool generation, so measure before assuming gains.
  • Document Intelligence and Knowledge Retrieval: Combine Semantic Search and Retrieval-Augmented Generation (RAG) to answer questions from contracts, policies, manuals, and support histories, with citations back to the source so answers can be verified.
  • Speech Recognition: Speech recognition uses AI algorithms to convert spoken language into text. You can use speech recognition to automate transcriptions, create voice-activated Assistants, or improve accessibility for users with disabilities.
  • Recommendation Engines: Hyper Personalization. Example Product, News, Content. AI-powered recommendation engines analyze user behavior and data to provide personalized recommendations. You can use recommendation engines to suggest products, services, or content based on a user's interests and preferences.
  • Predictive Analytics: Uses machine learning algorithms to analyze historical data and make predictions about future events. Examples - Sales forecasting, Inventory forecasting, Demand planning, optimize supply chain management
  • Recognition: Find an object or detect any pattern from any kind of data. Example Object detection, computer vision, Video/Image processing, voice recognition, and Facial Recognition. Image and video analysis use AI algorithms to recognize and interpret visual data. You can use this technology to automatically tag and organize images or videos, detect and track objects, or even identify faces for security or surveillance purposes.
  • Anomaly Detection: To find patterns from data and establish relation between those to find outliers. Example risk analysis
  • Fraud Detection: AI can help you detect and prevent fraud by analyzing patterns and anomalies in financial transactions. By leveraging AI-powered fraud detection tools, you can identify fraudulent activity in real-time and take appropriate action.
  • Automation: Autonomous cars/Drones, Autonomous operation. AI-powered automation can streamline and optimize business processes by automating repetitive or time-consuming tasks. For example, you can use AI-powered tools to automate customer support, financial analysis, or inventory management. Autonomous systems use AI to make decisions and take actions without human intervention. You can leverage autonomous systems to automate tasks such as driving, manufacturing, or monitoring and control systems.
  • Augmented Reality and Virtual Reality: AR and VR use AI to create immersive and interactive experiences. You can use AR and VR to create virtual product demonstrations, training simulations, or marketing campaigns.

Evidence check by pattern

Pattern Strongest current evidence Watch out for
Customer support assistance Field study of 5,179 agents: about 14% more issues resolved per hour, about 34% for novices Over-automating complex cases; customer frustration with dead-end bots
Coding assistants Mixed: a Google trial found about 21% faster; METR found experts 19% slower on familiar code in early 2025 Perception gap; security flaws; review burden
Writing, analysis, and consulting tasks BCG experiment: large gains inside the capability frontier Confident errors on tasks outside the frontier
Agentic automation Widespread experimentation, but scaling remains limited in any single function Compounding errors across steps; unclear accountability
Content creation Rapid adoption; freelance price pressure in writing and image work Demonetization of mass-produced content; labeling duties

Getting Started

A 90-day starter plan

  1. Weeks 1 to 2: Audit your work. List your recurring tasks and estimate hours for each. Pick three that are frequent, text-heavy, and easy to check.
  2. Weeks 3 to 4: Run small experiments. Try two or three different assistants on those tasks. Time yourself with and without AI on comparable work, and note every error you catch. Self-assessment alone is unreliable, so write the numbers down.
  3. Weeks 5 to 6: Learn the fundamentals. Take one beginner course from the list above and learn how models work at a high level, including why they hallucinate and how context affects answers.
  4. Weeks 7 to 9: Build one real thing. Create a reusable workflow: a prompt library with examples, a retrieval-based assistant over your own documents, or a small automation. If you code, use Python and an open model or API; if not, use codeless tools.
  5. Weeks 10 to 12: Package and share the result. Write up what you built, what it saved, and what failed. That write-up becomes a portfolio piece, an internal proposal, or the first case study for a service offering.

Load video
YouTube
Why Data Science Skill-Building Might Be Holding You Back | CareerCon 2019 | Kaggle
Dan Becker, Head of Kaggle Learn, makes a strong case against obsessive learning, and instead argues for a focused game plan for finding your first data science job. Hosted by Addison Howard. About Kaggle: Kaggle is the world's largest community of data scientists. Join us to compete, collaborate, learn, and do your data science work. Kaggle's platform is the fastest way to get started on a new data science project. Spin up a Jupyter notebook with a single click. Build with our huge repository of free code and data. Stumped? Ask the friendly Kaggle community for help.

Load video
YouTube
How to be Ready for Jobs in an Artificial Intelligence Driven World
The world is changing and is changing fast. Many of today’s traditional #jobs are at risk. With #AI becoming more predominant, how can we ensure that our children and their children can thrive in this future?

Load video
YouTube
How to get started in #AI? Introducing the Thumbtack Framework
Many people want to get started in AI but don't know where to begin. My advice is to start with your position of strength and leverage your background and skills. You don't all have to learn to program machine learning systems but you can use your past experience to position yourself in the future we're AI is going to be dominant. In this video I share a simple framework that will help you do that. I try to provide perspectives in AI and architecture - touching upon different topics. At times, I also help companies clarify their message in these domains. If you'd like to work with me, please send me a note (through my website www.drrajramesh.com) Thanks for watching my videos and for subscribing. www.drrajramesh.com www.linkedin.com/in/rajramesh

Load video
YouTube
Andrew Ng on Building a Career in Machine Learning
Title: Break Into AI: A Q&A with Andrew Ng on Building a Career in Machine Learning Date: 12/4/2018 Andrew Ng will share tips and tricks on how to break into AI. He will discuss some of the most valuable skills for today's machine learning engineers, how to gain the experience to successfully switch careers, and how to build a habit of lifelong learning. He will also take questions from aspiring engineers and business professionals who want to work on AI-powered products. SPEAKER Andrew Ng, General Partner, AI Fund; CEO, Landing AI; Adjunct Professor, Stanford University. Dr. Andrew Ng, a globally recognized leader in AI. As the former Chief Scientist at Baidu and the founding lead of Google Brain, he led the AI transformation of two of the world’s leading technology companies. A longtime advocate of accessible education, Dr. Ng is the Co-founder of Coursera, an online learning platform, and founder of deeplearning.ai, an AI education platform. Dr. Ng is also an Adjunct Professor at Stanford University’s Computer Science Department. MODERATOR Juan Miguel de Joya, UN ITU; ACM Practitioners Board is the in-house consultant for Artificial Intelligence and Emerging Technologies at the United Nations International Telecommunications Union.

Load video
YouTube
Machine Learning Engineer Jobs, Resume & Salary | Machine Learning Engineer Salary Report | Edureka
( Machine Learning Engineer Masters Program: https://www.edureka.co/masters-progra... ) This video will provide you with detailed knowledge of who is a Machine Learning Engineer, what are the salary trends, the job trends and the correct format of an ML engineer's resume. This video will also provide you with the job descriptions, roles and the skills required to become one successful ML Engineer.

Load video
YouTube
How to Build a Compelling Data Science Portfolio & Resume | Kaggle
Learn how to craft and tailor your Data Science resume to get noticed by Hiring Managers. Learn what to include, what not to include, and how to prioritize what to mention. About Kaggle: is the world's largest community of data scientists. Join us to compete, collaborate, learn, and do your data science work. Kaggle's platform is the fastest way to get started on a new data science project. Spin up a Jupyter notebook with a single click. Build with our huge repository of free code and data. Stumped? Ask the friendly Kaggle community for help.

Load video
YouTube
Data Works MD: Loyola Student Showcase - Predicting Outcomes and Profitability
Data Works MD June 2020 Predicting the Profitability of Out-of-Print Comic Books and Graphic Novels Like many collectibles, comic books and graphic novels have a wide price trajectory. While some are worth millions of dollars, most resell for far less than the purchase price. For a comic book business, the earlier you identify which items will rise or fall in price, the more you maximize your profits. The best course of action would be to purchase comics that are destined to appreciate in bulk from a distributor at wholesale prices. To do so, you need a way to identify which comics will appreciate in value. For my Loyola data science capstone project, I built a regression model that predicts how much a comic will appreciate when no longer available from the distributor.

Understanding and Predicting Frederick County Animal Shelter Outcomes According to the American Society for the Prevention of Cruelty to Animals (ASPCA), approximately 6.5 million animals enter animal shelters in the United States each year, roughly half dogs and half cats. Only half of these animals get adopted, and about 25% of them are euthanized every year. Although some animals may be euthanized because they are dangerous, there are many innocent animals that lose their lives because they could not find a home. For my final capstone project in Loyola University Maryland’s data science program, I conducted a thorough data analysis for Frederick County Animal Control (FCAC). I used a combination of data visualization and predictive analytics to help FCAC with their daily operations and potentially improve animal adoption rates. Specifically, I created models aimed at predicting whether or not an animal will be adopted. In this talk, I will walk through the end-to-end process of my project, including data cleaning, exploratory data analysis, predictive modeling, and interpreting the results. Greg first got into data and programming while studying astronomy as an undergrad at Yale University. During his time there he was also a member of the Varsity Heavyweight Crew. After graduation, Greg put astronomy on hold to pursue a career in rowing and for the past 7 years has been coaching rowing at the D1 level at Loyola University Maryland. While at Loyola, Greg has returned back to programming and data through the Master's in Data Science program. Greg will be graduating at the end of August and is actively looking for opportunities in the data science field. Kaitlyn graduated from UMBC in 2016 with a Bachelor’s Degree in Mathematics and recently completed her Master’s in Data Science from Loyola University Maryland. She has been working at Booz Allen Hamilton as a data scientist for the past four years.

Load video
YouTube
How to become a Full-Stack Deep Learning Engineer by Forrest Iandola
Historically, AI has been an algorithm-centric field. However, with the rise of Deep Neural Network (DNN), it is now the case that (1) large-scale data, (2) novel DNN models, and (3) efficient software and hardware infrastructure, are all key to success. The best outcomes often come from teams who understand the "full stack" from low-level hardware for DNNs to high-level applications of DNNs. Full-stack DNN teams are able to make big-picture tradeoffs in the development of data, models, and infrastructure, leading to practical solutions that exhibit unprecedented levels of accuracy, speed, energy-efficiency. In this talk, Forrest Iandola focuses on three main topics. First, he inclusively defines the "full stack" of skills and technologies that go into DNN engineering. Second, he describes a playbook for managers who want to build, coach, and grow a full-stack DNN engineering team. This playbook draws on lessons that Forrest have learned first-hand at UC Berkeley, Microsoft Research, and DeepScale. Finally, he provides advice on how a generalist or specialist engineer can engage with a full-stack DNN engineering team, and describes a path for how to ultimately become a full-stack DNN engineer. Forrest Iandola completed a PhD in Electrical Engineering and Computer Science at UC Berkeley, where his research focused on Deep Neural Networks. His advances in scalable training and efficient implementation of Deep Neural Networks led to the founding of DeepScale, where he is CEO. DeepScale is focused entirely on building perception systems for automated vehicles, and DeepScale has a number of engagements with automakers and automotive suppliers.

Roadmaps

Overall, roadmaps can be a valuable tool by providing a clear and concise overview of the learning journey. Treat any roadmap as a compass rather than a checklist: the field moves fast enough that the specific tools change every year, while fundamentals such as statistics, evaluation, and clear problem framing last.

Machine Learning (ML) Roadmap

Load video
YouTube
2020 Machine Learning Roadmap
Getting into machine learning is quite the adventure. And as any adventurer knows, sometimes it can be helpful to have a compass to figure out if you're heading in the right direction. Although the title of this video says machine learning roadmap, you should treat it as a compass. Explore it, follow your curiosity, learn something and use what you learn to create your next steps. Links: Interactive Machine Learning Roadmap - https://dbourke.link/mlmap

Large Language Model (LLM) Roadmap

The following subjects are derived from Ryan Nguyen's excellent 'How AI Built This' article, "This is what I'd do if I could learn how to build LLM from scratch where he asks the question, "What if start from absolute zero, knowing what I know now? Where would I begin? How would I tackle each challenge?" ...

Additions for 2025 and 2026. Since the original roadmap was written, several topics have become essential for anyone building with LLMs:

  • Reasoning models. Models that generate extended internal reasoning before answering. Learn when the extra cost and latency are worth it.
  • Tool use and the Model Context Protocol. How models call functions and connect to data sources through MCP servers; see the Model Context Protocol documentation and the Hugging Face MCP Course.
  • Agents and orchestration. Planning, memory, multi-step execution, and when a simple workflow beats a fully autonomous agent; see Agents/Assistants and the Hugging Face Agents Course.
  • Context engineering. Deciding what instructions, documents, examples, and tool results go into the context window, in what order, and what to leave out.
  • Evaluation. Building test sets from real tasks, scoring outputs automatically and with human review, and tracking regressions as models change.
  • Security and guardrails. Prompt injection, data leakage through tools, permission scoping for agents, and output filtering.
  • Cost and latency management. Choosing model sizes, caching, batching, and routing simple requests to cheaper models.

Load video
YouTube
Load video
YouTube

How to Get into AI

Load video
YouTube

Load video
YouTube
How to get into AI + Facebook AI reading list (the best resources)
Hi, I'm Oleksii and I work at Facebook AI Research. The list of the best resources to learn ML and AI is right here. 

 ========== Stage 1 ==========

-Theory: Linear Algebra, Statistics, Theory of Probability (any courses/books will work)

-Practice: Basic Python skills, NumPy 

 ========== Stage 2 ==========

-Theory: Online courses / books on ML and Deep Learning; Machine Learning course by Andrew Ng, Coursera; Deep Learning Book, Goodfellow; Stanford CS231n course; Recent MIT course: https://introtodeeplearning.com/; https://www.fast.ai/

-Practice: Pandas, PyTorch, Tensorflow and "official" tutorials (form their websites) 

 ========== Stage 3 ==========

-Theory: "Classic" papers, e.g. AlexNet, ResNet, BERT; Reading lists: https://deeplearning.net/reading-list/ https://github.com/ujjwalkarn/Machine... https://github.com/floodsung/Deep-Lea... MILA reading list: https://docs.google.com/document/d/1I... Andrej Karpathy's blog: https://karpathy.github.io/

-Practice: Random tutorials on the internet, Advanced tutorials, replicating papers 

 ========== Stage 4 ==========

-Theory: New paper on your topic, conferences, ArXiv, mailing lists and news channels; 

https://distill.pub/;

Sebastian Ruder: https://newsletter.ruder.io/ ; THE BATCH: https://www.deeplearning.ai/thebatch/

-Practice: Working on toy-projects; Contributing to open-source projects; Getting an internship; Trying to get hands-on experience in any way possible 

 ========= Never Stop =========

These all will be useful for anyone, engineer or researcher. Also try to concentrate on one particular topic and get an expert in it first, it will be much easier to accomplish and will open you opportunities right away. Don't try to become an expert in everything and also do not neglect "not prestigious" experience. Any experience is very valuable.

Mental Resilience

Building your "mental resilience" as you head into a world where you will need to continue to reinvent yourself to leverage/compete with artificial intelligence - continually making yourself relevant. To do so, you need to gain data science expertise; again an understanding how best leverage artificial intelligence/machine learning capabilities and applications.

The current evidence gives this advice a concrete shape. PwC found that the skills employers want in AI-exposed jobs are changing about 66 percent faster than in other jobs, so the ability to keep learning matters more than any single skill. Stanford's payroll research suggests the pressure falls hardest on early-career workers whose value rests mainly on codified knowledge that AI can reproduce, while experience, tacit knowledge, and judgment remain protective. And METR's developer study is a reminder that feeling more productive is not the same as being more productive, which argues for calm, deliberate measurement over anxiety-driven tool hopping. Practical habits that help: schedule regular learning time, build relationships with people who use AI well, keep a record of measured wins, and focus on problems where you bring context that a model cannot get on its own. See Psychology - Mental Health and productivity.

Load video
YouTube
Why the rise of AI makes "mental resilience" so important
Israeli historian Yuval Noah Harari explored the past and future of humanity in his books "Sapiens" and "Homo Deus." They became international bestsellers and were praised by a wide-range of thought leaders, including former President Barack Obama and Bill Gates. In his new book, "21 Lessons for the 21st Century," Harari focuses on the present and dissects the most pressing issues facing humanity. He joins "CBS This Morning" to discuss why it's important to teach children "emotional intelligence" and "mental resilience" as they head into a world where they will need to continue to reinvent themselves to compete with artificial intelligence.

Watch me Build a ...

Note: In 2019, Siraj Raval publicly acknowledged plagiarizing parts of a research paper and faced complaints about his paid machine learning course. His videos remain useful as quick demonstrations of building an AI app end to end, but verify technical and financial claims independently, and treat automated-investing demos such as the one below as teaching examples, not trading strategies.

Load video
YouTube
Watch Me Build an AI Startup
I'm going to build a medical imaging classification app called SmartMedScan! The potential customers for this app are medical professionals that need to scale and improve the accuracy of their diagnoses using AI. From ideation, to logo design, to integrating features like payments and AI into a single app, I'll show you my 10 step process. I hope that by seeing my thought process and getting familiar with the sequence of steps I'll demonstrate, you too will be as inspired as I am to use this technology to do something great for the world. Enjoy!

Load video
YouTube
How to Start an AI Startup
How are you supposed to get in on the AI hype? Deep Learning has enabled a whole new breed of applications, and there are still so many different opportunities to apply it in fields that are completely untapped. I'll go through the steps you need to take to start your own AI startup using a combination of my own experiences and best practices from the industry as a guide. From data collection to model training to picking a problem, we'll try to understand this challenging task.

Load video
YouTube
Make Money with Tensorflow 2.0
I've built an app called NeuralFund that uses Tensorflow 2.0 to make automated investment decisions. I used Tensorflow 2.0 to train a transformer network on time series data that i downloaded using the Yahoo Finance API. Then, I used TensorFlow Serving + Flask to create a simple web app around it. I'll explain what the important parts you should know in Tensorflow 2.0 are, then I'll guide you through my code & thought process of building an AI startup using it. Enjoy!

Load video
YouTube
7 Ways to Make Money with Machine Learning
Machine Learning is an amazing technology, but how are you supposed to earn a living from it? In this video, I'll break down 7 ways that anyone can earn money from anywhere in the world using machine learning. Well start by taking a look at whats called the "AI Value Chain" to learn who is currently making money in machine learning so that we can better chart out where we can contribute to the space. From startups, to competitions, to writing books, we've got a lot to cover in this video. Enjoy!

Load video
YouTube
How to Do Freelance AI Programming
You can build a sustainable full-time income from doing freelance AI programming work. In this video, i'm going to show you the steps you can take to start your journey as a freelancer. Whether you're a student or are employed full-time, you can begin the process of planning out a freelance career today. Getting clients, leveling up your skills, marketing yourself, setting up your financials, tools to help optimize your workflow, these are all aspects of the freelance life that i'll explain from my own personal experience.

Load video
YouTube
How to Make Money as a Programmer in 2018
I'll go through 5 methods that you can use to make money as a programmer! We are lucky in that our skill will only get more valuable to society over time. Links to everything I've discussed are below.

Learning Approaches

The most effective learners in this field combine three things: a small amount of theory at the right time, a lot of hands-on building, and honest feedback on results. The talks below argue for code-first, application-focused learning, which matches what employers increasingly test for. Two additions for the AI-assistant era: use AI as a tutor that explains, quizzes, and critiques your work rather than as a shortcut that does it for you, and periodically solve problems without assistance to confirm the skill is actually yours.

Load video
YouTube
Learning AI and ChatGPT isn’t that hard

Load video
YouTube
How to Learn Deep Learning (when you’re not a computer science PhD)
Talk video from meetup April 11, 2017 at AWS office in SF. Huge thanks to Amazon for providing venue, food/drink, and video recording! Abstract: Many people claim that Deep Learning needs to be a highly exclusive field, saying that you must spend years studying advanced math before you even begin to attempt it. Jeremy Howard and I believed that this was just not true, so we set out to see if we could teach Deep Learning to coders (with no math prerequisites) in 7 part-time weeks. Our students are now using Deep Learning to identify chainsaw noise in endangered rain forests, create translation resources for Pakistani languages, reduce farmer suicides in India, diagnose breast cancer, and more. We wanted to help them get results fast, so we taught them in a code-centric, application-focused way. I’ll share what we learnt about how to learn Deep Learning effectively, so that you can set out on your own learning journey. Rachel Thomas

Load video
YouTube
Learn Machine Learning in 3 Months (with curriculum)
How is a total beginner supposed to get started learning machine learning? I'm going to describe a 3 month curriculum to help you go from beginner to well-versed in machine learning. Its an accelerated learning plan, something i'd create for myself if I were to get started today, but I'm going to open source it for you guys. This curriculum will cover all the math concepts, the machine learning theory, and the Deep Learning theory to get you up to speed with the field as fast as possible. If anyone asks how to best get started with machine learning, direct them to this video!

Load video
YouTube
How to Study Machine Learning
Let me show you the techniques I use to study machine learning in this video. That includes living a healthy lifestyles, optimizing your learning environment, creating a personalized learning path, prioritizing effectively, and being an active learner. I'll demo the FAST technique, which you can use to help learn faster and more efficiently. I made this with machine learning technology in mind, but these techniques can be used for any field. Enjoy!

Resources

Additional Viewing

Load video
YouTube
Andrej Karpathy: Software Is Changing (Again) (Y Combinator AI Startup School, June 2025, about 40 minutes). Karpathy, a founding member of OpenAI and former head of AI at Tesla, argues that software has entered a third era in which natural language becomes a programming interface and LLMs act like a new kind of operating system. He emphasizes partial-autonomy products with fast human verification loops rather than fully autonomous agents. Useful for understanding where new products and jobs are forming, and why human review stays central. Watch on YouTube

Load video
YouTube
Andrew Ng: Building Faster with AI (Y Combinator AI Startup School, June 2025, about 44 minutes). Drawing on AI Fund, the venture studio he leads, Ng explains why execution speed predicts startup success, how agentic workflows change what small teams can build, and why concrete product ideas beat vague ones. He argues that AI coding tools shift the bottleneck from engineering to product decisions and user feedback, which is directly relevant to anyone building an AI business. Watch on YouTube

Load video
YouTube
How I Use LLMs, Andrej Karpathy (February 2025, about 2 hours 11 minutes). A practical, general-audience tour of the LLM ecosystem: choosing models and pricing tiers, when to use reasoning models, web search and deep research, file uploads, code interpreters, AI coding tools, voice, and image and video features. It is one of the best single resources for the "use AI better in your current job" path. Watch on YouTube

Load video
YouTube
Deep Dive into LLMs like ChatGPT, Andrej Karpathy (February 2025, about 3 hours 31 minutes). Karpathy walks through how chat models are built, from pretraining data and tokenization to fine-tuning and reinforcement learning, and explains the resulting "psychology" of models, including hallucinations and uneven abilities. Understanding these mechanics makes it much easier to predict where AI will help or fail in your own work. Watch on YouTube

References

Adoption and business value

Productivity research

Jobs, skills, and freelancing

Building blocks and learning

Consumer protection and regulation

Retrieved from "https://primo.ai/w/index.php?title=How_do_I_leverage_Artificial_Intelligence_(AI)%3F&oldid=46489"