Twelve months of AI agents and AI governance

The 48-week plan

Every course, book and certification the plan names, on its scheduled window. A monthly agent re-checks each one and flags anything that has been superseded — it flags, it never edits the plan.

Portfolio

One agent and one governance memo per shipped week.

Relevancy log

On the 1st of each month an agent re-checks every unfinished resource against the current state of the field, and looks for anything new the plan should know about. Its guardrail is that it flags — it never edits the curriculum.

1. September 2026

   2026-09-01

   The curriculum is largely current — most courses (Hugging Face, IBM, DeepLearning.AI, LangChain Academy) and certifications (IAPP AIGP v2.1) reflect 2026 best practices with no superseding editions. Two items are flagged "watch": the NIST GenAI Profile, which is being supplemented by new CAISI agent standards work, and the EU AI Act risk-classification guide, which needs updating for the July 2026 Digital Omnibus deadline changes even though its core framework still holds.

   Flagged

   • watchNIST AI Risk Management Framework + Generative AI Profile — NIST CAISI's February 2026 AI Agent Standards Initiative is producing new agent-specific guidance (with an Interoperability Profile due Q4 2026), so the core GenAI Profile remains valid but is being actively supplemented.
   • currentAgentic Artificial Intelligence — Bornet et al. — Independent 2026 roundups still rank it "best overall," co-authored by senior practitioners drawing on real agent deployments, with no newer edition superseding it.
   • currentHuman Compatible — Stuart Russell — It remains a foundational AI-safety text with no indication of a superseding edition or major reframing of its arguments.
   • currentHugging Face AI Agents Course — Updated Aug 8, 2026, and still called one of the best free starting points available in 2026 for hands-on agent framework learning.
   • currentIBM: Building AI Agents and Agentic Workflows — It teaches building agents with LangGraph, CrewAI, and AutoGen using OpenAI, Meta Llama, and IBM Granite models, matching the current mainstream agent-framework stack.
   • currentThe Alignment Problem — Brian Christian — No evidence of a superseding edition; it remains a standard reference on alignment history and concepts.
   • currentIAPP AIGP Body of Knowledge v2.1 — IAPP released version 2.1 effective February 2, 2026, which is exactly the version named in the resource title.
   • currentWeapons of Math Destruction — Cathy O'Neil — Remains a widely cited foundational text on algorithmic bias with no indication of being superseded.
   • currentDeepLearning.AI: Agentic AI (Andrew Ng) — It is current content that DeepLearning.AI updates regularly, and remains its flagship agentic AI offering in 2026.
   • currentIBM: RAG and Agentic AI Professional Certificate — Still listed among the best online courses for agentic AI alongside IBM's other current agentic workflow specialization.
   • watchEU AI Act practitioner risk-classification guide — The Digital Omnibus (Regulation (EU) 2026/1744), in force since 27 July 2026, postponed high-risk deadlines to Dec 2027/Aug 2028, so guides not reflecting this shift are outdated on timelines even though the underlying risk-classification architecture is unchanged.
   • currentIAPP AIGP Certification Exam — The exam now runs on AIGP Body of Knowledge version 2.1 (updated Feb 2, 2026), matching the current listed BoK.
   • currentUntangling AI — Matt Kesby — It is described as the best AI agents book published in 2026 and the only mainstream-publisher 2026 release in this category aimed at non-technical executives.
   • currentLangChain Academy (LangGraph) — It continues to expand its offerings, including new courses like Deep Agents, keeping the LangGraph flagship course current with the framework's evolution.
   • currentStanford AI Index 2026 — Responsible AI chapter — Published by Stanford HAI in April 2026 detailing AI's rapid advancement versus governance capacity, it is the latest edition and still current.

   Suggested additions

   • Check Point disclosed 11 vulnerabilities spanning LangChain, LangGraph, CrewAI, AutoGen, Microsoft Agent Framework, and Google ADK where prompt-controlled content can escalate into trusted framework logic, so learners should add adversarial/security testing to any agent-framework module.
   • A critical remote-code-execution flaw in IBM's Langflow (CVE-2026-9198) was added to CISA's Known Exploited Vulnerabilities catalog after active attacks, meaning curricula covering low-code agent builders must now include patch-management and secure-deployment practices.
   • EU AI Act Article 50 transparency obligations and GPAI enforcement powers took effect August 2, 2026, while the high-risk system deadline was pushed to December 2027, so learners need to know exactly which AI Act provisions are live now versus delayed.
   • Microsoft shipped Agent Framework 1.16.0 with configurable first-task wait timeouts, richer OpenTelemetry/OTLP observability, and core safety improvements, which learners building production agents should adopt for better monitoring and reliability.
   • Industry and regulators (Google's Agent Payments Protocol, NIST agent-identity concept work, and the proposed AI AGENT Act) are converging on verifiable, task-bounded authorization records for agents that handle money or legal actions, a governance pattern learners should practice before deploying transactional agents.

2. August 2026

   2026-08-16

   This curriculum's core textbooks and practitioner courses remain largely solid and current, though the Hugging Face course is starting to show gaps (e.g., no MCP coverage) as agent engineering matures. The governance/regulatory resources need closer attention: the EU AI Act guide is affected by the July 2026 Digital Omnibus that pushed back high-risk compliance deadlines, and the NIST AI RMF ecosystem is expanding with new agentic-specific and critical-infrastructure profiles still in draft, while the IAPP AIGP materials and Stanford AI Index are already fully up to date with their 2026 versions.

   Flagged

   • currentHuman Compatible — Stuart Russell — Foundational AI-safety text from Stuart Russell that remains a standard reference; no newer edition or replacement has emerged as of 2026.
   • currentAgentic Artificial Intelligence — Bornet et al. — This is a relatively new book (2025 release) with no revised edition; still the go-to practitioner guide for business-focused agentic AI, drawing on research across 167 companies according to author accounts.
   • watchHugging Face AI Agents Course — Agent stacks moved from prototype loops to operational systems expecting retries, traceability, state, and safer tool execution, and reviewers note the course has no MCP coverage, a growing gap in 2026 agent engineering.
   • watchNIST AI Risk Management Framework + Generative AI Profile — The AI RMF 1.0 is being revised under the White House AI Action Plan, with a new critical-infrastructure concept note released April 2026 and separate agentic-specific and interoperability profiles still in development for Q4 2026, so the resource is usable but the surrounding ecosystem is actively expanding.
   • currentIBM: Building AI Agents and Agentic Workflows — No evidence of supersession; IBM continues to actively maintain its agentic AI course offerings alongside its broader Coursera catalog in 2026.
   • currentIAPP AIGP Body of Knowledge v2.1 — Approved on 9 September 2025 and effective February 2, 2026, this Version 2.1 supersedes version 2.0.1 and is exactly the version this resource already reflects.
   • currentThe Alignment Problem — Brian Christian — Remains a widely-cited foundational text on AI alignment with no newer edition or direct replacement identified.
   • currentWeapons of Math Destruction — Cathy O'Neil — Still a standard reference on algorithmic harm and bias; no update or supersession found.
   • currentDeepLearning.AI: Agentic AI (Andrew Ng) — Remains a current flagship offering teaching iterative, multi-step agentic AI workflows, though one reviewer noted it is an introductory course that doesn't cover everything needed for production-grade systems.
   • currentIBM: RAG and Agentic AI Professional Certificate — The program has been updated to span generative AI fundamentals, RAG, vector databases, multimodal applications, and multi-agent orchestration using LangChain, LangGraph, CrewAI, AutoGen, BeeAI, and the Model Context Protocol, keeping it aligned with current production patterns.
   • watchEU AI Act practitioner risk-classification guide — The Act was amended by the Digital Omnibus on AI, in force since 27 July 2026, postponing high-risk requirements to 2 December 2027 and 2 August 2028 while adding two new prohibitions effective 2 December 2026, so a guide predating these changes would have outdated timelines and classification carve-outs.
   • currentIAPP AIGP Certification Exam — The exam now runs on the updated AIGP Body of Knowledge 2026, effective Feb 2, 2026, and this resource already reflects that update.
   • currentStanford AI Index 2026 — Responsible AI chapter — Now in its ninth edition and produced by an interdisciplinary steering committee, this is the freshly released 2026 edition and remains the most comprehensive independent snapshot of AI's state.
   • currentUntangling AI — Matt Kesby — This is a newly published 2026 title with no prior edition to supersede it, focused on enterprise automation and AI agent adoption strategy.
   • currentLangChain Academy (LangGraph) — LangGraph remains the industry standard for building agentic workflows, and LangChain Academy continues to be regarded as an authoritative flagship source for learning LangGraph agent architectures in 2026.

   Suggested additions

   • The EU AI Act's high-risk provisions, including risk management, human oversight, and conformity assessment, became enforceable on August 2, 2026, alongside transparency rules requiring chatbots to self-identify and synthetic media to carry labels/watermarks, so any 2026 curriculum on agent deployment must now treat these as live legal obligations, not future requirements.
   • Amazon renamed Bedrock Agents to "Bedrock Agents Classic," closing it to new customers as of July 30, 2026 and freezing its model catalog, meaning learners should pivot to AgentCore, which remains a separate, framework-agnostic runtime with dedicated services for runtime, gateway, memory, identity, and observability.
   • Microsoft's Agent Framework shipped version 1.14.0 with broader model support, including a Mistral chat client, workflow checkpoint resume, and new hooks and storage options across core and Foundry, an important update for anyone building on Microsoft's converged AutoGen/Semantic Kernel stack.
   • Real-world production guidance has shifted from toy benchmarks toward failure-mode testing, since practitioners now stress that meaningful framework comparisons must test what happens when tools fail, APIs timeout, or context windows overflow rather than relying on documentation claims.
   • Recently reported safety incidents showing agentic misalignment in leading AI systems (flagged in early-August 2026 agent-news roundups) underscore why any 2026 curriculum should now pair framework skills with incident-response and oversight practices, not just building tutorials.

3. August 2026

   2026-08-16

   The curriculum remains largely solid, with most courses and books actively maintained or recently published. The governance-related resources (NIST RMF, IAPP AIGP exam, EU AI Act guide) need the closest attention due to incoming agentic-specific standards, updated exam content, and the Digital Omnibus's 2027 deferral, while Weapons of Math Destruction is aging relative to the generative-AI era it now sits alongside.

   Flagged

   • currentHugging Face AI Agents Course — The course anchors on agent fundamentals, covers multiple frameworks (smolagents, LlamaIndex, LangGraph), includes a benchmarked final project, and remains actively maintained into 2026.
   • currentHuman Compatible — Stuart Russell — No successor or supersession was found; it remains a foundational, framework-agnostic text on AI safety philosophy.
   • watchNIST AI Risk Management Framework + Generative AI Profile — NIST CAISI's new AI Agent Standards Initiative and planned Q4 2026 Interoperability Profile, plus an ongoing AI RMF 1.0 revision, mean a new agentic-specific layer is coming even though the current GenAI Profile remains in active use.
   • currentAgentic Artificial Intelligence — Bornet et al. — No evidence of a newer edition or competing resource surfaced, so it remains current.
   • currentIBM: Building AI Agents and Agentic Workflows — The specialization covers building agents using LangGraph, CrewAI, AG2 (AutoGen), and BeeAI with hands-on labs, matching the current 2026 agent-framework landscape.
   • currentIAPP AIGP Body of Knowledge v2.1 — Version 2.1, effective February 2, 2026, kept the four-domain structure with targeted refinements, and this is the version the resource already reflects.
   • currentThe Alignment Problem — Brian Christian — No evidence of a superseding text was found; it remains a widely cited conceptual survey of alignment research.
   • watchWeapons of Math Destruction — Cathy O'Neil — The 2016 book predates the generative-AI/agentic era and its examples are increasingly dated, though its core bias framework is still referenced.
   • currentDeepLearning.AI: Agentic AI (Andrew Ng) — The course still teaches building agentic AI systems through iterative, multi-step workflows and remains a live, certificate-bearing course.
   • currentIBM: RAG and Agentic AI Professional Certificate — The certificate continues to run on Coursera with hands-on labs in generative, retrieval-augmented, and agentic techniques, with no replacement program found.
   • watchIAPP AIGP Certification Exam — The 2026 exam now emphasizes complete AI systems, adding content on AI provider roles, Fundamental Rights Impact Assessments, agentic AI architectures, and third-party risk, requiring updated v2.1 materials.
   • watchEU AI Act practitioner risk-classification guide — The Digital Omnibus's provisional deferral of the Annex III high-risk deadline to 2 December 2027 and the Commission's new draft classification guidelines mean a guide not reflecting these changes would misstate current rules.
   • currentStanford AI Index 2026 — Responsible AI chapter — The 2026 edition is the 9th installment with Responsible AI as a core chapter, and it is the newest edition available with no successor yet.
   • currentUntangling AI — Matt Kesby — Published February 17, 2026, it is one of the newest resources in the curriculum with no newer edition or replacement found.
   • currentLangChain Academy (LangGraph) — LangChain Academy continues offering self-paced courses, including new content on Deep Agents, showing active maintenance and expansion beyond the original LangGraph module.

   Suggested additions

   • The Model Context Protocol released its largest-ever spec revision (2026-07-28), making MCP stateless with new extensions, Tasks, and OAuth-aligned authorization—curriculum modules on agent tool-calling/interoperability need updating to this version.
   • The EU's Digital Omnibus (Regulation (EU) 2026/1744, in force July 27, 2026) pushed high-risk AI obligations (Annex III) to December 2027 and Annex I to August 2028, so learners should stop treating "August 2026" as the high-risk compliance deadline in their study materials.
   • Despite that delay, the EU AI Act's Article 50 transparency rules (AI self-disclosure, synthetic-media labeling) and the AI Office's enforcement powers over GPAI providers took effect exactly as scheduled on August 2, 2026, making this the first live enforcement milestone governance learners should track.
   • Amazon rebranded Bedrock Agents to "Bedrock Agents Classic," closing it to new customers as of July 30, 2026, while pushing all new development to the separate AgentCore runtime—agent-framework curricula built on Bedrock Agents should pivot examples to AgentCore.
   • Microsoft Agent Framework shipped v1.14.0 with broader model support (including a Mistral chat client) and workflow checkpoint/resume, showing continued rapid maturation post-GA that structured courses referencing MAF should keep in sync with.

4. August 2026

   2026-08-15

   This curriculum's resources are largely holding up well — most courses (Hugging Face, IBM, DeepLearning.AI, LangChain Academy) have been actively refreshed to reflect 2026 agent-stack norms, and the classic governance/ethics books remain unsuperseded. The main areas to watch are regulatory-timeline resources: the EU AI Act risk-classification guide and IAPP AIGP exam both sit atop frameworks that shifted materially in 2026 (the Digital Omnibus delay and the BoK v2.1 update), so practitioners should verify they're using post-update editions.

   Flagged

   • currentHugging Face AI Agents Course — The course remains actively maintained and anchors on agent fundamentals before covering multiple frameworks (smolagents, LlamaIndex, LangGraph), keeping pace with 2026 agent-stack norms.
   • currentAgentic Artificial Intelligence — Bornet et al. — No new edition or replacement has emerged since its 2025 release, and it remains a widely cited business-oriented guide to agentic AI implementation.
   • currentHuman Compatible — Stuart Russell — This remains a foundational, unrevised AI safety text with no indication of being superseded by a newer canonical work on the same core alignment thesis.
   • watchNIST AI Risk Management Framework + Generative AI Profile — The core RMF and GenAI Profile remain the reference standard, but NIST's CAISI AI Agent Standards Initiative and its planned Q4 2026 AI Agent Interoperability Profile mean agent-specific guidance is still emerging on top of the base framework.
   • currentIBM: Building AI Agents and Agentic Workflows — The course continues to be refreshed with current tooling, teaching students to build next-generation AI agents using LangGraph, CrewAI, and AutoGen with hands-on labs on OpenAI GPT, Meta Llama, and IBM Granite models.
   • currentThe Alignment Problem — Brian Christian — No newer edition has replaced it, and it remains a standard reference for AI alignment concepts even as agentic-specific risk discussions have expanded around it.
   • currentIAPP AIGP Body of Knowledge v2.1 — This is the presently effective version, updated to v2.1 on February 2, 2026, and it remains the four-domain structure used to test candidates today.
   • currentDeepLearning.AI: Agentic AI (Andrew Ng) — The course remains DeepLearning.AI's active flagship agentic offering, though a reviewer noted it is introductory and other sources are needed for production-grade agentic AI systems.
   • currentWeapons of Math Destruction — Cathy O'Neil — The book remains a widely referenced foundational text on algorithmic bias and has not been superseded, though it predates today's specific regulatory frameworks like the EU AI Act and NIST RMF.
   • currentIBM: RAG and Agentic AI Professional Certificate — The certificate was recently expanded and refreshed into ten courses spanning generative AI fundamentals, RAG, vector databases, multimodal applications, and multi-agent orchestration using LangChain, LangGraph, CrewAI, AutoGen, BeeAI, and MCP.
   • watchEU AI Act practitioner risk-classification guide — The Digital Omnibus amendment (in force since 27 July 2026) postponed high-risk requirements, with Chapter III now applying from 2 December 2027, so any guide not reflecting these deferred deadlines and narrowed "safety component" definitions is outdated in its timeline details.
   • watchIAPP AIGP Certification Exam — The exam now tests against the updated BoK v2.1, but since the official IAPP textbook is not expected until early summer 2026, there is a temporary study-material gap.
   • currentStanford AI Index 2026 — Responsible AI chapter — This is the latest edition, now in its ninth year, produced by an interdisciplinary steering committee, remaining the most current authoritative annual snapshot.
   • currentUntangling AI — Matt Kesby — This is a newly released 2026 title with no indication of a subsequent edition or replacement, still positioned as a practical roadmap for business leaders assessing AI adoption maturity.
   • currentLangChain Academy (LangGraph) — The academy remains the authoritative, actively expanding source, now also teaching newer capabilities like building long-running agents with Deep Agents, LangChain's open-source agent harness.

   Suggested additions

   • The Model Context Protocol shipped its 2026-07-28 spec finalized on July 28, making the core stateless, deprecating sessions/roots/sampling, and introducing a Tasks extension—learners need to update MCP-based agent tooling for breaking changes.
   • The EU AI Act's high-risk provisions (risk management, human oversight, conformity assessment) and transparency/watermarking rules became enforceable on August 2, 2026, making this the first governance milestone any 2026 curriculum must cover in practice, not just theory.
   • AWS renamed Bedrock Agents to "Bedrock Agents Classic," closing it to new customers on July 30, 2026 and freezing its model catalog, so learners should pivot to the newer AgentCore runtime for current-generation agent projects.
   • Microsoft Agent Framework jumped to 1.13.0 in early August with reusable session stores, expanded telemetry/observability, and stronger workflow replay/approvals—relevant for anyone building production-grade multi-agent workflows on Azure.
   • ISO/IEC 42001 certification is rapidly becoming a de facto vendor requirement, with major firms (Presidio, TechnipFMC, NeenOpal) certifying in July–August 2026, so governance learners should prioritize this standard over less commercially recognized credentials.