Welcome back! Gartner expects 40% of agentic AI projects to be scrapped by 2027, and the roadmap below shows exactly where they die. Then the vocabulary, the learning order and a booster choice you have been making on autopilot. 

In today's Generative AI Academy Newsletter:

  • That’s your ultimate agentic AI roadmap: 100 steps updated for November 2026, and the three most teams skip.

  • 20 agentic AI terms, decoded: from tool calling to MCP, A2A and agent evals.

  • Learn GenAI in the right order: three levels from Python to production, and the one jump to avoid.

  • Stop picking your gradient booster by habit: XGBoost, LightGBM and CatBoost differ in two places only.

That's your ultimate agentic AI roadmap

Updated for November 2026. Save it.

The 10 steps haven't changed. The tools on them have:

→ GPT-4o is off the board. OpenAI's GPT-6 family replaced it
→ AutoGen is now part of Microsoft Agent Framework 1.0, merged with Semantic Kernel
→ MCP is now how agents plug into tools. It's run by the Linux Foundation's Agentic AI Foundation, which passed 300 members in October

Agents also left the demo stage.

McKinsey's 2026 survey: 40% of large companies are scaling AI agents, up from 27% a year ago.

But Gartner still expects over 40% of agentic AI projects to be scrapped by the end of 2027, mostly over cost and unclear value.

The fix is already on the board: step 5 (design) and step 8 (monitor) come before step 10 (scale).

Most teams invert that. Demo working in a week, evaluation skipped because nothing looks broken, then at 10,000 runs nobody can explain a failure or name the cost per task.

Three items where projects die:

→ 41. Define objective. A goal you cannot score is a goal you cannot debug.
→ 55. Test workflows. A failed agent is 12 model calls you cannot untangle afterward.
→ 76. Cost monitoring. Cost per task stays invisible until volume arrives, and by then the architecture is set.

Four numbers decide it: task success rate, cost per task, hallucination rate, error rate.

If you cannot produce all four for your agent today, you are still at step 5.

We are raising to build applied AI on top of this.

Master Claude Code in 6 sessions

You already know the developers who seem to move twice as fast as everyone else.

They're not typing faster. They're: 

  • Orchestrating teams of agents

  • Running parallel Claude instances across git worktrees

  • Wiring MCPs into their whole stack 

  • Scheduling loops that keep working while they sleep

Almost none of it is written down in one place.

Mastering Claude Code course puts it all in one place.

6 live sessions, each one hands-on:

  • You start with the agentic loop and model selection

  • You build Skills and full agent teams from scratch

  • You apply Boris Cherny's 13 tips plus 6 new April 2026 techniques

  • You set up all 3 tiers of CLAUDE.md as a second brain that gets smarter every session

  • You wire MCPs into Slack, GitHub, Gmail and Supabase

  • You leave having built, tested and shipped your own custom Skill as a capstone

This is the deep track, the habits that separate casual users from power users. 

Pricing is locked in for early sign-ups.

20 Agentic AI terms, decoded

Building an agent is more than an LLM with a few tools attached. Behind a working one sits planning, memory, routing and evaluation.

Fundamentals

  1. AI Agent → uses models and tools to reach a goal

  2. Tool Calling → connects the model to APIs and functions

  3. ReAct → reason, act, observe, repeat

  4. Planner → splits a goal into steps

  5. Reflection → reviews its own output and revises

Memory and execution

  1. Agent Memory → carries information across interactions

  2. Agentic RAG → retrieves knowledge as the task needs it

  3. State → holds context through a run

  4. Router → sends a request to the right agent

  5. Orchestrator → coordinates work across agents

Multi-agent

  1. Multi-Agent System → specialized agents working together

  2. Handoff → passes control from one agent to another

  3. Checkpointer → saves progress so a failed run can resume

  4. Human-in-the-Loop → human approval before a consequential action

Security

  1. Guardrail → enforces operational limits

  2. Prompt Injection → instructions smuggled in through untrusted input

  3. Sandbox → isolates code execution from production

  4. MCP → one standard for connecting apps to tools

  5. A2A → lets independent agents talk to each other

Reliability

  1. Agent Evals → measures accuracy, tool use and task completion

Shipping an agent is one job. Shipping one that holds up is understanding how these 20 fit together.

At GenAI.Works, we're helping businesses navigate this shift by connecting AI innovation with the people and companies ready to adopt it. 

Learn GenAI in the right order

Learning GenAI feels overwhelming for one reason.

Nobody tells you what to learn first. LLMs, RAG, agents, MCP, fine-tuning, deployment. All of it at once, in no order.

Here is the order.

Level 1: Foundations

→ Python, JSON, HTTP
→ Tokens and context windows
→ Prompt engineering
→ APIs and function calling
→ Structured outputs

You can build a working tool after this level.

Level 2: Real applications

→ Embeddings and vector databases
→ RAG, chunking, reranking
→ FastAPI
→ Evaluation and observability
→ Guardrails and access control

Here your app answers from your own documents.

Level 3: Production systems

→ Agents and ReAct
→ Memory and state
→ MCP and multi-agent setups
→ LoRA and QLoRA fine-tuning
→ Quantization and inference optimization

This is where cost per request and reliability under load start to matter.

One mistake to skip: jumping to agents before you can evaluate a single model call. An agent that fails is 12 model calls you cannot debug.

Build the foundation. Ship something with your own data. Then move up.

The goal is not collecting AI skills. It is becoming an engineer whose AI applications hold up.

GenAI Works opened its round to everyone: from $1,000, with up to 20% bonus shares.

We run go-to-market for NVIDIA, Oracle, Google and 300+ AI brands through a 14M+ community, and revenue grew 2.5x last year.

Learn AI without paying for a single tool

You already know the people who seem to get twice as much done with AI. They're not on better subscriptions. They're:

  • Running research, analysis and transcription through free tiers

  • Writing prompts with frameworks instead of guessing

  • Building custom agents they reuse every week

  • Shipping actual tools, not just chatting with a model

Almost none of it needs a paid plan.

AI in Practice puts the whole path in one place. 7 modules, 19 lessons, 11+ hours, every tool free or free-tier:

  • You start with how LLMs actually work, tokens, attention, context windows, RLHF, and where they break

  • You write and iterate prompts using established frameworks, context engineering and meta-prompting

  • You build a QR code generator, an email-reply bot, a Zapier form automation and an appointment-booking agent

  • You work in ChatGPT, Gemini, Claude, Perplexity, Manus, GitHub Copilot and Zapier

  • You cover GDPR, the EU AI Act, bias and why most corporate AI projects fail

  • You leave with a portfolio website presenting everything you built as the capstone

Taught by Larno Visser, who trains teams at Philips, BDO, municipalities and schools.

Built for zero-coding backgrounds. Early bird is $60 instead of $80, and the offer closes in 5 days.
→ Save your seat

Stop picking your gradient booster by habit

Most data scientists pick XGBoost, LightGBM or CatBoost out of habit, then spend hours tuning.

The three only diverge in two places. Knowing which two makes the choice obvious.

First, how each tree grows.

XGBoost grows level by level. Predictable, harder to overfit, slower.

LightGBM grows the best leaf first. Fastest by a wide margin, and it overfits small data.

CatBoost splits one feature per level. Rigid, and very fast at prediction time.

Second, how they read categorical columns.

XGBoost splits on value sets. Native support, no smoothing, so rare values overfit.

LightGBM splits on integer codes, but only if you flag the column. Forget the flag and it reads your categories as numbers.

CatBoost encodes using only past rows, so the encoding cannot leak. Nothing to configure.

Which one to pick.

Mostly numeric columns and if you plan to tune deeply, use XGBoost. Widest ecosystem, safest default in production.

Millions of rows, or iterating fast, use LightGBM. Cap the leaves or it memorizes.

Lots of categorical columns, especially high cardinality, use CatBoost. Strongest defaults, least tuning to reach a real number.

Btw, early rounds in AI companies usually go to VCs. GenAI Works opened its round to everyone with up to 20% bonus shares.

We run go-to-market for NVIDIA, Oracle, Google and 300+ AI brands, and revenue grew 2.5x last year.

Everything else you shouldn't miss

  • AI chatbots send shoppers to fake store sites at very different rates: a study of 6,795 answers found risky websites in 15% of Perplexity’s responses, 8% of ChatGPT’s and 4% of Gemini’s, so the citation is the part worth checking before you click.

  • Cell type is the variable that was blurring genomics predictions: AlphaGenome 2 predicts one cellular state at a time instead of dozens at once, and found 30% of causative mutations in immune cells against 16% for the previous version.

  • Cryptic crosswords broke AI because the clue is built to mislead: Darmok solved all seven championship puzzles with no errors by running each clue through three models and filling in the confident answers first, using those letters as a head start on the rest.

  • Your AI assistant inherits the uptime of everything it connects to: ChatGPT and Claude both stayed online through Microsoft’s 10-hour outage and both lost access to your mail and calendar, because the connector was what failed rather than the model.

Learn more about AI from the experts building it

Follow us on Instagram for fast, visual AI updates in 30 seconds. 

Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business. 

📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.

🐦 Follow us on X for breaking AI news and real-time industry updates.

Learn how to build your next AI application with practical resources and expert guidance.

Explore investment opportunities in the future of AI and join our community-backed growth journey.