ctaio.dev Ask AI Subscribe free

AI · AI Education · agentic engineering resources

Best Resources to Learn Agentic Engineering in 2027

The fastest way to learn agentic engineering in 2027 is to pair Simon Willison's free Agentic Engineering Patterns guide with Anthropic's Claude Code best practices and the Model Context Protocol documentation, then add evals training from Hamel Husain and Shreya Shankar. Latent Space, the AI Engineer conferences and the Agentic AI Foundation keep that foundation current. Agentic engineering is the discipline of building and running AI agents, coding agents included, reliably enough that a team can depend on them: scoping tasks, building context, wiring tools through MCP, supervising runs and evaluating what comes back. I've grouped the 13 resources below by format so a technology leader can pick one per format instead of trying to read everything.

By · Published November 19, 2026 · Checked November 19, 2026

Engineer's desk at night with terminal sessions on two monitors and an agent workflow sketched on a whiteboard

Key takeaways

  1. Start with two free documents: Simon Willison's Agentic Engineering Patterns guide and Anthropic's Claude Code best-practices page. Together they cover scoping, testing, subagents, hooks and permission rules in an afternoon of reading.
  2. Evals are the skill most teams skip and the one that decides whether agents survive production. Hamel Husain and Shreya Shankar's Maven course costs US$4,500, with the next cohort running 30 Jan–27 Mar 2027; Husain's free guides on hamel.dev cover the basics.
  3. Read the MCP documentation before you approve any agent that touches internal systems. MCP is now a project of the Linux Foundation's Agentic AI Foundation, which runs AGNTCon + MCPCon North America in San Francisco on 28–29 Apr 2027.
  4. Buy at most one structured course per engineer. Free options from Anthropic cover MCP and Claude Code, and our separate agentic AI courses comparison covers 13 longer programmes.
  5. One feed is enough to stay current: Latent Space for interviews and the weekday AINews digest, or the AI Engineer YouTube channel, at 1.04M subscribers, for conference talks.

Which resources are worth your time for learning agentic engineering?

Ranked within each group by how much a technology leader gets out of it. "Checked" says where each number came from and when.

Start here: guides and documentation

  1. Free online guide · newsletter 73K+ subscribers

    For: Staff engineers and engineering managers who want one vendor-neutral reference for working with coding agents.

    A free guide he keeps extending, covering how coding agents work, Git with agents, subagents, red/green TDD and agentic manual testing, with the prompts he uses in an appendix. It treats Claude Code and OpenAI Codex even-handedly and opens by separating agentic engineering from vibe coding. His newsletter carries the running commentary the guide is built from.

    Checked: Creator site (guide contents) and Substack page (73K+ subscribers), 19 Nov 2026

  2. Free · 29 articles as of Nov 2026

    For: Architects and AI leads who want first-hand write-ups from a lab that runs agents at scale.

    Anthropic's engineering team writes up how it builds, evaluates and deploys agents, from the December 2024 'Building effective agents' patterns (prompt chaining, routing, orchestrator-workers, evaluator-optimizer) to a March 2026 piece on harness design for long-running coding work and a November 2026 piece on agent memory. Read 'Building effective agents' first, and note Anthropic's own caveat that much of the tooling it describes has changed since.

    Checked: Vendor engineering blog, 19 Nov 2026

  3. Free official documentation

    For: Teams rolling out Claude Code, and anyone drafting an internal coding-agent policy.

    The most practical official document on running a coding agent day to day: write specific prompts, keep CLAUDE.md short, give the agent a check to run, enforce rules with hooks, limit access with permission rules and the sandbox, and know what subagents cost. Most of the advice carries over to other coding agents even where the commands differ.

    Checked: Vendor documentation, 19 Nov 2026

  4. Free · current spec version 2026-07-28

    For: Platform and security engineers who will build MCP servers or sign off on them.

    MCP is the open standard that lets agents reach tools and data, supported by clients including Claude, ChatGPT, VS Code and Cursor. The docs cover architecture, building servers and clients, and a security best-practices tutorial that belongs on every reviewer's reading list. The protocol is now listed as a project of the Agentic AI Foundation.

    Checked: Project documentation site and aaif.io project list, 19 Nov 2026

Practitioners to follow

  1. 543K YouTube subscribers · newsletter 2–3 times a week · AINews every weekday

    For: Engineers and AI leads who want to know which agent tools and techniques practitioners picked up this month.

    swyx's publication bills itself as 'The AI Engineer publication' and pairs long technical interviews with a weekday AINews digest. Agent tooling gets serious discussion here early, usually with the people who built it. Pick the podcast if you commute and AINews if you want a daily scan.

    Checked: Publisher site and About page; YouTube channel page (543K), 19 Nov 2026. The 200,000+ audience figure in the channel description is the publisher's own claim

  2. Free blog · 182 posts

    For: Engineers designing evals and LLM product systems, and leaders who want patterns tested in production.

    Yan is a Member of Technical Staff at Anthropic and was a Principal Applied Scientist at Amazon, and his posts turn production experience into reusable patterns. Recent writing covers building cybersecurity evals (a sandboxed target, tools and a grader) and how to work and compound with AI. Each post is short enough to discuss in a team reading group.

    Checked: Creator site and About page, 19 Nov 2026

  3. Blog, talks and workshops

    For: Engineers experimenting with long-running autonomous coding loops.

    Huntley's July 2025 post 'Ralph Wiggum as a software engineer' named the technique of running a coding agent in a loop until the spec is met, and the idea travelled far enough to get VentureBeat coverage. His blog is the primary source for the method and its limits. Read it alongside a cost and stop-condition policy before anyone points a loop at a real repository.

    Checked: Creator site (post dated 14 Jul 2025) and contact page, 19 Nov 2026

  4. 154K subscribers · Claude Code Deep Mastery playlist

    For: Hands-on engineers who learn best by watching complete agentic coding sessions.

    The channel says its aim is to help viewers 'evolve into Agentic engineers', and the videos are long screen-shares of Claude Code and rival coding agents on real builds, including head-to-head comparisons. He sells paid courses at agenticengineer.com; the free videos are enough to judge whether the style suits your team.

    Checked: YouTube channel page, 19 Nov 2026

Courses

  1. US$4,500 · live cohort, next 30 Jan–27 Mar 2027 · 4.7/5 from 1,058 reviews

    For: Engineers and product managers who own an agent or LLM feature in production and cannot yet say whether it is getting better.

    The best-known evals course: 33 live sessions across 11 lessons, all recorded, with homework, 10+ hours of office hours and a 200+ page course reader. Husain says more than 5,000 students from 500+ companies have taken it. If the price is too steep for a first step, his free AI Evals Guide and blog on hamel.dev cover the core method.

    Checked: Course page on Maven (price, dates, rating) and instructor site (student count), 19 Nov 2026

  2. Free · self-paced

    For: Engineers new to MCP or Claude Code who need a structured, zero-cost starting point.

    Anthropic's free catalogue includes Introduction to Model Context Protocol (servers and clients in Python, built on tools, resources and prompts), Model Context Protocol: Advanced Topics, and Claude Code in Action, which teaches plan mode, compaction, hooks and verification for long hands-off sessions. It is the cheapest way to give a whole team the same baseline in a week.

    Checked: Course catalogue and course pages (marked FREE), 19 Nov 2026

Book

  1. Book · print, Kindle and O'Reilly platform

    For: Engineering leaders who want durable principles under the fast-changing tools.

    The standard reference on building applications on foundation models, with a practical framework for developing and deploying them that covers models, datasets, evaluation benchmarks and application patterns. Huyen says it has been the most-read book on the O'Reilly platform since release. It predates many current coding-agent tools, so use it for principles and the documentation above for current mechanics.

    Checked: Author site, 19 Nov 2026

Communities and events

  1. 1.04M YouTube subscribers · Europe 17–19 Feb 2027, London · World's Fair 29 Jun–1 Jul 2027, San Francisco

    For: Engineers who build agents for a living, and leaders who want the talks without the travel.

    A practitioner conference series that turns its in-person events into free videos on a channel with over a million subscribers. AI Engineer Europe runs at ExCeL London and the World's Fair at Moscone West, with smaller AIEi events in Sydney, Miami and Singapore. Send the engineers who will run agents in production, and have everyone else watch the recordings.

    Checked: Organiser site (dates, venues) and YouTube channel page, 19 Nov 2026

  2. Linux Foundation · AGNTCon + MCPCon North America 28–29 Apr 2027, San Francisco

    For: Platform leads and architects who care where agent standards are heading.

    The Linux Foundation body that hosts MCP, Agent2Agent, AGENTS.md and goose, so its working groups are where agent standards get settled. mlops.community now redirects to it, and its community hub runs weekly online Coding Agents Lunch & Learn sessions. AGNTCon + MCPCon Europe follows in London on 15–16 Sep 2027.

    Checked: Foundation site (projects, events calendar) and community hub, 19 Nov 2026

What should an engineering leader learn first about agentic engineering?

Learn supervision before prompting. Our agentic engineering guide splits the practice into four parts: scoping tasks, building context, supervising and recovering agents while they run, and reviewing their output. Generating a change is the easy part now. Spotting an agent that has stalled silently, or one that is confidently wrong, is where teams lose time and trust.

That framing sets the reading order. Start with Simon Willison's patterns guide and the Claude Code best-practices page, because both focus on scoping, checks and review. Then read the MCP documentation, since tool access is where an agent stops being a chat window and starts acting on your systems. Evals come third, through Hamel Husain's free material or the paid course, because without them you cannot tell whether any change to prompts, models or tools helped.

Leave the broader theory for last. Chip Huyen's book and the Anthropic engineering archive give you the vocabulary to argue about architecture, but they will not tell your team what to do on Monday. The order matters because scope creep and confident errors are among the recurring failure modes our guide lists, and scope and verification are the first two topics here.

Which resources fit which role on your team?

Match the resource to the job, because one reading list for everyone wastes most people's time. Senior engineers who will run coding agents daily get the most from Willison's guide, IndyDevDan's long screen-shares and Claude Code in Action. Platform and security engineers should own the MCP documentation and its security tutorial, and should read our MCP security guide before the first internal server goes live.

Engineering managers and product managers who own an agent feature need evals more than anything else. That is the case for the Maven course, or for Eugene Yan's posts if the budget is not there. CTOs, CAIOs and VPs of Engineering need enough fluency to set policy and judge vendor claims: Latent Space, the Anthropic engineering blog and two or three AI Engineer talks a month will do that in under two hours a week.

The staffing point is easy to miss. The role agentic engineering needs is an engineer with operational instincts, and scoping and supervision skills often sit in different people. Pick resources that build both, and do not hand all of them to whoever was labelled the prompt engineer.

How do you turn these resources into team capability?

Tie the learning to one real workflow, or it stays reading. I'd pick a single repository and a narrow, repeatable task such as dependency upgrades or test backfill, run a six-week reading group on Willison's guide and the Claude Code docs, and have the group write the team's agent policy as its output. The policy should cover permissions, review gates and when a run must stop.

Build the eval set in the same six weeks. Our guide to testing and evaluating AI agents shows how to start with a few dozen real cases instead of a benchmark. Add stop conditions and budgets from the agentic loops guide before anyone tries a long-running loop of the kind Geoffrey Huntley describes.

Measure the result against a baseline you trust. The agentic engineering guide cites a 2025 METR trial in which experienced open-source developers took about 19% longer with AI tools while believing they were faster. Your own numbers will differ, which is the reason to collect them before scaling the workflow to other teams.

How much time and budget should you set aside?

Most of this list is free, so engineering time is the real cost. I'd budget two or three focused days for a senior engineer to work through Willison's guide, the Claude Code docs and the MCP documentation, plus a day or two for Anthropic's free courses. Put that time in the sprint plan, or delivery work will squeeze it out.

Money matters for three things: an evals course for whoever owns agent quality, conference travel for one or two engineers a year, and the tokens burned in practice runs. The Maven course is US$4,500 per seat, and conference ticket prices vary by tier, so check the organiser sites before you budget. For longer, credentialed programmes from MIT, Berkeley or Cambridge, our agentic AI courses comparison lists 13 options with prices; I would not pay for a second course from this page on top of one of those.

Track practice token spend from the first week. Our 90-day Claude Code cost log shows what one heavy user's bill looks like over time, which is a better planning number than any list price.

What can you safely skip?

Skip anything that teaches one framework's API without the operating discipline around it. Framework courses age within months, and the patterns in this list outlast them. Skip certificates as a hiring signal too: they show someone finished a course and say nothing about whether they can recover a stalled agent without losing its work.

Treat vendor material with care even when it is good. Anthropic's own note on 'Building effective agents' says much of the tooling has changed since December 2024, and the same will be true of every document here within a year. Check dates, prefer the primary docs over summaries, and revisit the list each quarter.

Finally, skip the agent where a script will do. Our guide on when not to use AI agents lists four conditions that rule them out. Teaching a team to say no to an agent is part of agentic engineering, and it saves more money than any course on this page.

Listed here, or should be?

If your entry is out of date or you think something belongs on this list, email [email protected]. Entries are editorial and free. A featured, labelled Sponsored slot is available through vendor listings.

Related questions

More from AI Education

Frequently Asked Questions

What is the best free resource to learn agentic engineering?
Simon Willison's Agentic Engineering Patterns guide is the best free starting point because it is vendor-neutral and covers how coding agents work, Git workflows, subagents and testing. Pair it with Anthropic's free Claude Academy courses on MCP and Claude Code for structured practice.
Is the AI Evals course by Hamel Husain and Shreya Shankar worth US$4,500?
It is worth it for a team that already runs an agent or LLM feature in production and cannot measure whether changes help. The course has 33 recorded live sessions, homework and a 200+ page reader, and holds a 4.7/5 rating from 1,058 reviews on Maven. Teams earlier in the journey should start with the free guides on hamel.dev.
How is agentic engineering different from vibe coding?
Vibe coding means building software by prompting AI tools without reviewing the code they write. Agentic engineering is the discipline of running agents reliably in production, with scoped tasks, evals, permission rules and human review. See our vibe coding resources page for the lighter end of the spectrum.
Which conference should engineers attend to learn agentic engineering in 2027?
AI Engineer Europe (17–19 Feb 2027, London) and the AI Engineer World's Fair (29 Jun–1 Jul 2027, San Francisco) are the most practitioner-heavy options, and their talks are free on YouTube. AGNTCon + MCPCon North America (28–29 Apr 2027, San Francisco) suits teams focused on MCP and agent standards.
Do I need to learn MCP to do agentic engineering?
Anyone who builds or approves agents that reach internal tools or data should understand MCP, because it is the standard most agent clients use for that access. The official documentation and Anthropic's free Introduction to Model Context Protocol course cover the basics in a few hours.
·
Thomas Prommer
Thomas Prommer Technology Executive — CTO/CIO/CTAIO

These salary reports are built on firsthand hiring experience across 20+ years of engineering leadership (adidas, $9B platform, 500+ engineers) and a proprietary network of 200+ executive recruiters and headhunters who share placement data with us directly. As a top-1% expert on institutional investor networks, I've conducted 200+ technical due diligence consultations for PE/VC firms including Blackstone, Bain Capital, and Berenberg — work that requires current, accurate compensation benchmarks across every seniority level. Our team cross-references recruiter data with BLS statistics, job board salary disclosures, and executive compensation surveys to produce ranges you can actually negotiate with.