AI already writes the code. Give your team the system to direct it.
Six live sessions for your team to build its AI development harness: the system that lets you delegate work to agents and trust the result. When it ends, the team takes it away ready to apply to your projects.
Up to €997 per person · the larger the group, the lower the per-person price
- developers trained
- 10,000+
- years building software
- 15+
- Google Developer Expert
- GDE
- of AI Expert, live
- 6th cohort
- 90 student testimonials
- ★★★★★
Where you may have seen us
From loose prompts to a working system
The same request, two very different outcomes. The difference isn't the model or better prompting: it's the environment around the agent.
make me Google login
→ reading AGENTS.md and context
→ spec: scope + acceptance criteria
→ implementing (only files in scope)
✓ tests 12/12 · lint ok
✓ PR ready for your review
Turn the harness on
- tests
- ✓ 12/12
- scope
- only what the spec says
- your role
- reviewing the PR
That environment has a name: harness engineering. In this training your team builds it, with your stack and your rules.
Six sessions, six layers. A harness ready for your projects
Each session adds a layer on top of the previous one, with examples from your stack. Scroll down and watch the team harness come together.
session 1 · before writing code
Discovery
How models really work (context, cost and limits), and a discovery skill that forces the open questions of a feature to be answered before anyone implements it.
session 1 — project
$ skill discovery "export orders"
→ who uses it? which formats? which permissions?
+ docs/research/export-orders.md
✓ scope settled before touching code
session 2 · shared context
Context and specs
The team’s rules, architecture and conventions move into the repository. Every feature starts with a spec that says what gets built and when it is done, for people and agents alike.
session 2 — AGENTS.md
# AGENTS.md
- Architecture: docs/architecture.md
- Specs in /specs before implementing
✓ specs/export-orders.md · acceptance criteria
session 3 · permissions and roles
Environments and control
Choosing tools with judgment, with cost and privacy on the table, and putting limits on them: permissions, hooks and separate subagents to implement and to validate.
session 3 — .agents/
$ agent run implementer specs/export-orders.md
→ implementer: only files in scope
→ validator: read-only, against the spec
✓ pre-commit hook: lint + tests
session 4 · parallel, with evidence
Loop engineering
From issue to PR: several agents in parallel, each in its own worktree, using the team’s skills and MCPs. Nothing counts as done until build, lint and tests are green.
session 4 — team workflow
$ git worktree add ../export-orders
→ 3 agents in parallel, no collisions
✓ build · lint · tests 48/48
✓ PR #212 ready for review
session 5 · APIs, RAG and local models
AI in your product
AI goes into the product with technical control: APIs, structured output, function calling, RAG, evals, and local models when privacy requires them.
session 5 — src/ai/search.ts
await rag.search("return policy")
→ 4 relevant chunks · embeddings
→ evals: 18/20 answers with the right source
✓ answer citing the document
session 6 · the complete system
Adoption
Everything together, from spec to merge, with human review where it matters. And a plan to roll it out to the team’s other projects without depending on a single person.
session 6 — adoption
✏️ export-orders → spec
⚙️ search-filters → implementing
👀 sso-login → awaiting review
✅ docs/harness.md · next project
✓ harness complete
Tools will change. The system stays yours.
The team ends with a shared harness designed to live in the repository: it works with Codex, Claude Code, Cursor, Copilot or OpenCode, and does not depend on whoever set it up.
What your team takes away when the training ends
These are the pieces the team builds during the training, ready to take to your repositories the next day:
- ├── AGENTS.md
Shared context
Rules, architecture and conventions written once, for the whole team and for any agent.
- ├── specs/
Specs before execution
Every feature with scope and acceptance criteria. You review what was agreed, not what the agent came up with.
- ├── .agents/skills/
Repeatable processes
Tests, migrations, refactors or releases turned into skills the whole team uses, instead of one person’s tricks.
- └── definition-of-done.md
Review criteria
The evidence required before merging: build, lint, tests, and human review where the risk calls for it.
For teams that already build software
It does not matter whether the team barely uses AI or has spent months with agents by trial and error. What it needs is to know how to build software. This is what it will find (and what it won’t):
What your team covers in each session
The basis for working with judgment, and the first piece of the harness: pinning down a feature before anyone, person or agent, starts coding.
- LLM foundations: tokens, context, cost, limits and reliability.
- AI-assisted research: research workflows over documentation, repositories and new domains.
- Prototyping: seeing an idea before investing in building it.
- Discovery skill: the questions to answer before implementing.
Working with agents is no longer about writing good prompts, it is about managing context. We build the project’s context layer and a spec workflow the team can adopt.
- Context engineering:
AGENTS.md, architecture, rules and project conventions. - Spec-Driven Development: from idea to verifiable feature, with or without tools like OpenSpec or Spec-Kit.
- Specs as a contract: what is agreed is what gets built, for people and agents.
- Continuity: memory and context handoff between sessions and between people.
The tool landscape, to choose with judgment, and the operating practices that let agents execute without turning the project into a black box.
- Environment types: IDE, ADE and CLI: Cursor, Claude Code, Codex, OpenCode or Copilot, with cost and privacy on the table.
- Control layers: rules, hooks, permissions, plan mode and working modes.
- Subagents with roles: implementation and validation kept apart.
- Safe operation: restore points and review before integrating.
From using AI to orchestrating it in the team’s real workflow: loops the agents repeat (plan, implement, verify, fix) and several agents moving forward at once.
- Complete flow: spec → issues → implementation → PRs ready for review.
- Safe parallelism: worktrees and GitHub workflows without collisions.
- Team skills: turning repeatable processes into shared skills.
- MCPs and Definition of Done: external tools and closing with evidence.
Bringing AI into the product with technical control, from the quick test to an integration that holds up in production.
- AI APIs: providers, SDKs, tokens, streaming and cost control.
- Advanced capabilities: structured output, function calling, RAG, embeddings and evals.
- Local models: LM Studio, Ollama, privacy, latency and real limits.
- Real integration: observability and validation in production.
The harness working end to end, and how to extend it to the rest of the team and its projects without depending on a single person.
- The harness as an engine: agents using context, specs and skills on real work.
- Human oversight: review where it matters, automation where it helps.
- Adoption: which project to start with and how to extend it.
- Next steps: how to keep improving the harness with every new model.
Foundations and discovery
The basis for working with judgment, and the first piece of the harness: pinning down a feature before anyone, person or agent, starts coding.
- LLM foundations: tokens, context, cost, limits and reliability.
- AI-assisted research: research workflows over documentation, repositories and new domains.
- Prototyping: seeing an idea before investing in building it.
- Discovery skill: the questions to answer before implementing.
Two ways to start
To train a group, we run an edition just for your team. If you would rather start with one or two people and your team speaks Spanish, they can join the open edition.
Private edition
Up to €997
per person · the per-person price goes down as the group grows
Only your team, with dates, pace and examples adapted to your stack.
- 6 live sessions of 4 hours
- Remote or at your offices
- In English or Spanish
- Taught by Antonio and Nino, who split the topics
- Eligible for FUNDAE funding (Spain)
AI Expert open edition
€997
per person · starts 3 November
The full 6-week programme, with professionals from other companies, each working on their own project.
- Live classes and Q&A sessions
- Recordings and future editions included
- AI models included for 2 months
- Taught in Spanish
- Duration
- 24 h in 6 sessions of 4 h
- Format
- Remote or on-site
- Language
- English or Spanish
- Tools
- Whatever you already use
The final price depends on group size and format. After a first call we send you a fixed proposal, with no commitment.
What changes when the team stops asking and starts directing
Their own words, as a diff: in red how they worked before, in green what they took away. Nothing rewritten, only trimmed.
diff --git a/alumnos/santiago-perez-barber.md b/alumnos/santiago-perez-barber.md
@@ AI Expert · +19 −8 @@
1−You start out just "tinkering" with AI tools…
2+…end up mastering a set of resources that are useful for your daily work…
3+…and exponentially increase your productivity.
Santiago Pérez BarberBytacora Soluciones Informáticas SL · ★★★★★
And the rest of the history
90 reviews · ★★★★★ · every program
Hover to pause · click a review to read it in full
Instructors who apply AI in real projects
Antonio and Nino combine software development, agent systems, automation, data and teaching experience. They work directly on both training and company projects.
antonio@devexpert
Antonio LeivaGDE
DevExpert founder · Google Developer Expert
nino@devexpert
Nino RuanoAI · Big Data
AI and Big Data specialist · Instructor
It is a real shell (well, almost). Type help to see the commands.
Frequently asked questions
And if your question isn't here, write to us on WhatsApp and we'll answer you personally.
01How much does a private edition cost?+
02Can the training be subsidised?+
03Is it taught in English?+
04How does it fit into the team’s schedule?+
05Which tools will we use?+
06Do you work with our code?+
07How is it different from the open edition?+
08People on the team have very different levels with AI. Is that a problem?+
Your team’s harness starts with a conversation
Tell us how many you are, what you work with and what is getting in the way. We will send you a fixed proposal, with no commitment.