Media
Five products in parallel, with an agentic harness of its own.
One of Spain's largest media groups. We helped its teams ship several products at the same time working with AI, and left them an agentic harness to keep doing it without us.
- Client
- Vocento
- Service
- AI transformation of delivery
- First phase
- May – July 2026
- Now
- Phase 2 · Data, in progress
Nacho Corcuera · Diego San Cristóbal
Velchanos team on the project
- product streams in parallel
- 5product streams in parallel
- agentic harness handed to the teams
- 1agentic harness handed to the teams
- data projects in the second phase
- 4data projects in the second phase
Context
Several products to ship at once, and AI used ad hoc.
Vocento had several digital products in flight at the same time, with different teams. There was appetite to work with AI, but everyone did it their own way, with no shared method and no guarantees on quality or security.
The challenge
- Several products in motion at once, with different teams and owners.
- Scattered AI use, with no method shared between teams.
- Architecture, security and code quality as non-negotiables.
- Making sure the capability stayed in-house, not with the supplier.
What we did
A way to work in parallel, and a tool that stays in-house.
1
Parallel streams
Each product with its own owner, a shared weekly review and a channel per stream so teams never block each other.
2
Agentic harness
A plugin of skills, specialised agents, commands and workflows fitted to the way Vocento works.
3
The full cycle with AI
Explore, plan, implement and multi-agent review, with architecture and security built in as guardrails.
4
Autonomy
The harness stays with the teams. Today we are extending it to the Data area, across four new projects.
The agentic harness
The whole development cycle, with agents and guardrails.
Teams work with agents at every stage, always inside the house rules: architecture, security and automated reviews come built in. People decide what reaches production.
Technology
- Skills plugin
- Specialised agents
- Multi-agent review
- Google models (Vertex AI)
- CI/CD and GitOps
Working cycle with the harness
- Explore
- Plan
- Implement
- Multi-agent review
- Production
GuardrailsArchitecture · Security · Legacy code · CI/CD
- Skills
- house knowledge
- Agents
- specialised
- Commands
- for each task
- Workflows
- end to end
Phase 2 · Data: the same approach across four data projects, with the harness as the deliverable so the team stays autonomous.
Next case
Barcos Menorca