AI Workshop for developers.
Generating code faster is easy. The real challenge is continuing to ship reliable software without sacrificing quality.
26NovThursday · 2026
- Schedule
- 10:00–17:00
- Format
- Remote
- Seats
- Only 20


With Pablo, Rafa and Javi1,300+ devs have already taken the workshop
Wallapop
Leadtech
Kairos DS
Holded
Lambdaloopers
Kapsch
Edpuzzle
Vitro Bio
Saltoki
Capitole
Imatia
BBVA
Shakers
Promofarma
Volkswagen
S2 Grupo
Plexus
Voicemod
Avaibook
Intuix
Auctane
Vibia
Focus on the outer loop. Delegate the inner loop to the agent.
Your team already generates code with AI, but someone still reviews every PR by hand and can't keep up with so much code. It's bailing water out of a sinking boat with a bucket. The problem isn't the agent: there's no Agent Harness and no training to use it well.
Tools change. Processes stay.
Copilot, Cursor, Claude Code, Codex… The go-to tool changes every few months. That's why we don't teach you how to use a tool: we teach you how to build the Agent Harness that works with all of them.
- A single AGENTS.md and skills that any agent understands.
- Tests, hooks and CI that validate the code, no matter who writes it.
- Evals to switch models or agents based on data, not gut feeling.
A bug in production. A PR to review. Without anyone opening the IDE.
This is how the Self-Healing Software Loop we build in the workshop works. Scroll to see it step by step.
- 01 · Detect · 10:42:03
Sentry captures the error
A new exception in production. With its stack trace, the deployed version and the affected users.
Issues · productionNewTypeError: Cannot read properties of undefined (reading 'total')at calculateTotal (src/checkout/Checkout.ts:42)at CheckoutPage.render (src/checkout/CheckoutPage.tsx:18)at renderWithHooks (react-dom.js:1123)Events · last 5 min - 02 · Notify · 10:42:07
You get notified in Slack
Avoid notification overload by filtering only the ones that are really worth it.
Message #production-alerts👀 11 reply2 repliesTurbotuga is typing - 03 · Investigate and fix · 10:43:15
An agent finds the cause
An automation launches a cloud agent that replies in the thread: it reads the stack trace, searches the code, writes a test that reproduces the failure and fixes it.
Message #production-alerts👀 12 repliesTurbotuga is typing - 04 · Deliver · 10:46:12
The PR, ready for review
With the test that reproduces the bug, the fix and a green CI. You review and decide.
fix: handle empty cart in calculateTotal #482Conversation 1Commits 1Checks 3Files changed 2All checks have passed
3 successful checksCI / lint— Successful in 21sCI / test— Successful in 1m 12sCI / architecture— Successful in 38sReview required · Waiting for your review
TypeError in Checkout.ts:42 when the cart is empty. Adds a test that reproduces the bug.3 successful checks
The workshop doesn't stop here.
We go one step further and see how to bring AI workflows to more of your day-to-day tasks.
Every token that isn't cached, you pay for again.
Learn which model and effort to use for each task and how to order the context to take advantage of the cache. With Cache Explorer you'll see which part is reused in each iteration.
If the prefix doesn't change, each iteration only pays for what's new. If you break it, you pay for everything again. Reference price: $3 per million new tokens and $0.30 cached.
You don't need to ride every hype wave.
We try every new thing so you don't have to. Only what passes the filter makes it to the workshop. Starting now isn't arriving late: you skip the intermediate versions, the abandoned tools and the interface changes.
- A new model every week
- The IDE is dead
- Claude Design
- The magic prompt
- Yet another agent framework
- Vibe coding without tests
- MCP for everything
- Browser inside the terminal
- The UI changed again
- Benchmark record
- 100% autonomous agents
- 10x in a weekend
- Prompt caching
- Must-have extension
- The end of programming
- What we use every day at Codely
- What still works months later
- What saves time or money in a measurable way
- What works with any agent
Taught by Pablo, Rafa and Javi, from Codely.
After years focused on software architecture and design, we now apply it to coding with AI. What we teach is what we use every day at Codely.
for developers
Pablo
Rafa
Javi- ExplanationClear concepts, with live demos.
- ExercisesHands-on practice on a real project, in groups.
- ProTipsDay-to-day tricks you won't find in the docs.
Also available
for companies.
Align your team and get the most out of AI. Choose how:
AI Acceleration Pack: from where you are to where you want to be.
It's not just a day of training. It's complete support so AI stays in your team's day-to-day work.
- 01AI usage assessmentWe analyze how your team uses AI today: tools, harness, costs and workflow. You get a diagnosis and clear priorities.
- 02In-company workshopAn exclusive edition for your team, adapted to what came out of the assessment.
- 03Coaching sessionsRecurring sessions with AI Champions to review progress, remove blockers and bring the team up to date with what's worth it.
Choose how to live the workshop.
Next edition: November 26. There are only 20 seats and the price goes up as they sell out.
- 1 day · 6 live hours
- Remote and with 20 seats
- Exercises on a real project
- Everything in Individual for each person
- A single booking for the whole group
- Same workflow and criteria for the whole group
- Invoice in the company's name
- A single process for all enrollments
- Same workflow and criteria for the whole team
- Private edition
- Flexible date
- Flexible schedule
- You leave with shared criteria, debated only among yourselves
- Your team's questions, answered in private
- AI usage assessment
- Exclusive workshop
- Guidance sessions
- Diagnosis and priorities for your team
- A workshop to get everyone on the same page
- Recurring guidance sessions with AI Champions
Frequently asked questions
You don't need to have taken v1 to join v2. The difference is the focus: v1 was about getting the most out of the agent with AGENTS.md, progressive disclosure, testing, worktrees and the built-in browser. v2 focuses on the system around it:
- The outer loop: the system around the agent, with architecture tests, hooks and CI as guardrails.
- Self-healing: from a production error to a PR without opening the IDE.
- KTLO automations and up-to-date dependencies without supply chain attacks.
- The agent loop inside out, which model to pick and how to take advantage of the cache.
- Evals to measure whether your context works.
- An Agent Harness that works with any agent, and several agents in parallel.
And we dropped what didn't stand the test of time, like orchestrating subagents by role or using an MCP for everything.
TypeErrorinCheckout.ts:42when the cart is empty. Adds a test that reproduces the bug.