Here's the AI Roadmap You Actually Need
Tina HuangTina HuangKey takeaways
A useful AI learning roadmap starts with foundational skills, then branches into six paths: Strategist, Rogue, Alchemist, Crafter, Sage, and Machinist. The right path depends on your goals, budget, and learning preferences; the roadmap does not require choosing only one class.
Tutorial Island covers generative AI basics, prompting, and general-purpose tools; one foundational course path is enough to get started. The later paths range from AI leadership and agent-building to coding, model training, and hardware, with different levels of structure and resources; several areas, especially local AI and hardware, require substantial self-directed experimentation.
Video Flow
- 0100:00
The AI roadmap begins with shared foundations
The shared starter skills are understanding generative AI, prompting, and general AI tools. Several courses can cover them: the Google AI Professional Certificate is the preferred choice over Google AI Essentials plus Prompting Essentials for a broader, more interactive program, while Andrew Ng’s courses, Elements of AI, Anthropic courses, and OpenAI courses suit different budgets, learning styles, and vendor preferences.
- 0207:25
Strategist: lead AI adoption and set direction
Strategists compare vendors, shape human–AI teams, set organizational priorities, and lead adoption while balancing business measures such as ROI. The free, tool-agnostic OpenAI AI Leadership course is the recommended starting point for people without a strong vendor affiliation; AI for Everyone is an optional supplement, though it predates the generative AI boom.
- 0310:22
Rogue: experiment with agents and custom workflows
Rogues learn by building and testing agents, making useful workflows repeatable, and assembling custom setups; there is no single general course for this hands-on path. The starting choice is an agent ecosystem and its documentation, with options including Claude, OpenAI, Microsoft Copilot, Hermes, and Open Claude; the cohort-based Omachi option is mentioned without firsthand evaluation.
- 0413:37
Alchemist: build reliable agents for business workflows
Alchemists build agents, connect them to tools and knowledge, evaluate reliability, and deploy and monitor them within business workflows. The route can be no-code or code; either way, foundational learning should lead into a chosen platform and reliability practices, while the source identifies several courses, platforms, and books for those needs.
- 0516:59
Crafter: build and responsibly ship AI products
Crafters use AI coding tools to build AI-native apps and products, potentially without knowing how to code; their skills include directing coding agents, integrating AI into products, and shipping securely. There is no well-established, comprehensive foundational course, so tool-specific resources matter, while Beyond Vibe Coding and Anthropic’s AI-native SDLC playbook cover development practices and team workflows.
- 0619:33
Sage: understand models, then specialize
Sages study neural networks and transformers, then specialize in fine-tuning existing models or pre-training models from scratch; both tracks also require learning deployment. The suggested foundation options include building neural networks from scratch, a Stanford course, or Three Blue One Brown animations, while structured deployment resources are comparatively scarce.
- 0721:58
Machinist: combine AI with hardware
Machinists run AI locally on physical devices, study constraints such as memory, bandwidth, compute, and quantization, and may progress into data-center systems or robotics. The path combines practical tinkering with theoretical study: Machine Learning Systems offers free texts and labs, MIT’s TinyML course provides a university-style route, and NVIDIA and Hugging Face offer resources for deployment and robotics; the field has fewer learning resources and calls for experimentation.
- 0825:04
Choose a primary class, or combine interests
The six classes are not mutually exclusive: the creator identifies Alchemist as a primary path, with Crafter and Rogue as minor interests and Machinist as another area of interest. The roadmap ends with an invitation to identify a class and learning progress.
Tools & Resources
One of the two Coursera courses in the tool-agnostic Google Essentials starter option.
- 01One of the two Coursera courses in the tool-agnostic Google Essentials starter option.01:20
The second Coursera course in the Google Essentials starter option, covering prompting.
- 01The second Coursera course in the Google Essentials starter option, covering prompting.01:20
A broader, more interactive Coursera certificate covering practical AI uses and app building.
- 01A broader, more interactive Coursera certificate covering practical AI uses and app building.01:50
Andrew Ng’s introductory course, which can be paired with prompting study for the starter skills.
- 01Andrew Ng’s introductory course, which can be paired with prompting study for the starter skills.02:43
A prompting course in the Andrew Ng starter path; the source does not specify a more exact title.
- 01A prompting course in the Andrew Ng starter path; the source does not specify a more exact title.02:52
A free, extensive University of Helsinki course with interactive text and exercises.
- 01A free, extensive University of Helsinki course with interactive text and exercises.03:45
An Anthropic course in the Claude-focused starter curriculum.
- 01An Anthropic course in the Claude-focused starter curriculum.04:04
Anthropic course material covering AI fluency and foundational skills.
- 01Anthropic course material covering AI fluency and foundational skills.04:12
Anthropic course material on AI capabilities and limitations.
- 01Anthropic course material on AI capabilities and limitations.04:12
Prompting documentation included in Anthropic’s starter resources.
- 01Prompting documentation included in Anthropic’s starter resources.04:16
An OpenAI course in the self-paced, exercise- and assessment-based starter path.
- 01An OpenAI course in the self-paced, exercise- and assessment-based starter path.04:47
An OpenAI course that complements AI Foundations in the starter path.
- 01An OpenAI course that complements AI Foundations in the starter path.04:51
An additional OpenAI course that can be combined with its two foundations courses for a certificate.
- 01An additional OpenAI course that can be combined with its two foundations courses for a certificate.05:12
Recorded cohort-based intensive with project assignments and a capstone; no longer live or maintained.
- 01Recorded cohort-based intensive with project assignments and a capstone; no longer live or maintained.05:40
Free Microsoft and LinkedIn course with videos and a final exam; the discussed version was outdated, with an update then scheduled for September 25, 2026.
- 01Free Microsoft and LinkedIn course with videos and a final exam; the discussed version was outdated, with an update then scheduled for September 25, 2026.06:10
A copy-paste prompt in the companion resource to help match a learner’s goals and preferences to a course path.
- 01A copy-paste prompt in the companion resource to help match a learner’s goals and preferences to a course path.06:46
OpenAI’s self-paced course on governance, stakeholders, adoption roadmaps, and initial AI strategy.
- 01OpenAI’s self-paced course on governance, stakeholders, adoption roadmaps, and initial AI strategy.08:36
Andrew Ng’s free-to-audit course with enduring organizational frameworks, although its generative AI content predates the current boom.
- 01Andrew Ng’s free-to-audit course with enduring organizational frameworks, although its generative AI content predates the current boom.09:04
Ethan Mollick’s book on frameworks for thinking about AI in organizations.
- 01Ethan Mollick’s book on frameworks for thinking about AI in organizations.09:59
Arvind Narayanan and Sayash Kapoor’s skeptical account of what AI cannot do in organizations.
- 01Arvind Narayanan and Sayash Kapoor’s skeptical account of what AI cannot do in organizations.10:02
Anthropic’s agent ecosystem and documentation for building agent workflows.
- 01Anthropic’s agent ecosystem and documentation for building agent workflows.11:38
OpenAI agent tools and documentation for building agents and multi-agent workflows.
- 01OpenAI agent tools and documentation for building agents and multi-agent workflows.11:43
Microsoft’s agent and Copilot ecosystem, listed as an option for building workflows.
- 01Microsoft’s agent and Copilot ecosystem, listed as an option for building workflows.11:45
An open-source agent option listed for experimentation and workflow building.
- 01An open-source agent option listed for experimentation and workflow building.11:47
An open-source agent option listed as an alternative to vendor-specific ecosystems.
- 01An open-source agent option listed as an alternative to vendor-specific ecosystems.11:47
A cohort-based agent course or ecosystem option mentioned without firsthand use or evaluation.
- 01A cohort-based agent course or ecosystem option mentioned without firsthand use or evaluation.12:07
Serverless inference and fine-tuning for open models; the demonstration fine-tunes a model on personal study notes for reuse in applications.
- 01Serverless inference and fine-tuning for open models; the demonstration fine-tunes a model on personal study notes for reuse in applications.12:41
A recommended foundational course for learning how agents work on the no-code route.
- 01A recommended foundational course for learning how agents work on the no-code route.14:36
Vanderbilt course recommended as a foundational option for the no-code route.
- 01Vanderbilt course recommended as a foundational option for the no-code route.14:40
Andrew Ng’s video- and assignment-based agent course; the transcript does not give a more exact title.
- 01Andrew Ng’s video- and assignment-based agent course; the transcript does not give a more exact title.15:29
Interactive text-and-code learning resources with an emphasis on open-source AI.
- 01Interactive text-and-code learning resources with an emphasis on open-source AI.15:32
A cohort-based intensive listed as a code-route learning option.
- 01A cohort-based intensive listed as a code-route learning option.15:39
An agent-building platform whose resources can support implementation and reliability learning.
- 01An agent-building platform whose resources can support implementation and reliability learning.15:54
Learning resources covering agent reliability and related practices.
- 01Learning resources covering agent reliability and related practices.16:22
Chip Huyen’s book, recommended for learning about reliability in AI systems.
- 01Chip Huyen’s book, recommended for learning about reliability in AI systems.16:25
A cohort-based course option for learning reliability; no exact course title is supplied.
- 01A cohort-based course option for learning reliability; no exact course title is supplied.16:24
An AI coding tool listed as one possible Crafter development environment.
- 01An AI coding tool listed as one possible Crafter development environment.17:38
Anthropic’s coding tool, listed as a possible Crafter development environment.
- 01Anthropic’s coding tool, listed as a possible Crafter development environment.17:38
An AI coding editor listed as a possible Crafter development environment.
- 01An AI coding editor listed as a possible Crafter development environment.17:40
An AI coding assistant listed as a possible Crafter development environment.
- 01An AI coding assistant listed as a possible Crafter development environment.17:40
An AI app-building tool listed as a possible Crafter development environment.
- 01An AI app-building tool listed as a possible Crafter development environment.17:40
An AI app-building tool listed as a possible Crafter development environment.
- 01An AI app-building tool listed as a possible Crafter development environment.17:40
A technical book on sustainable building and deployment; some sections assume coding knowledge.
- 01A technical book on sustainable building and deployment; some sections assume coding knowledge.18:49
Anthropic’s Claude-oriented playbook for running software development teams and shipping AI products.
- 01Anthropic’s Claude-oriented playbook for running software development teams and shipping AI products.19:06
A prompt in the free companion resource to help identify a suitable coding tool.
- 01A prompt in the free companion resource to help identify a suitable coding tool.19:20
Andrej Karpathy’s course builds neural networks from scratch with code.
- 01Andrej Karpathy’s course builds neural networks from scratch with code.20:40
A Stanford course listed as an option for foundational model learning; the exact course title is not stated.
- 01A Stanford course listed as an option for foundational model learning; the exact course title is not stated.20:46
Visual animations listed as an option for learning model foundations.
- 01Visual animations listed as an option for learning model foundations.20:48
Training and certification resources for model deployment and AI infrastructure.
Two free textbook volumes with interactive labs and software, plus optional hardware kits, for learning machine-learning systems.
- 01Two free textbook volumes with interactive labs and software, plus optional hardware kits, for learning machine-learning systems.23:37
MIT university course with theory and assignments for efficient AI computing.
- 01MIT university course with theory and assignments for efficient AI computing.23:50
Free robotics learning resource for learners seeking a deeper introduction to robotics.
- 01Free robotics learning resource for learners seeking a deeper introduction to robotics.24:29