Your team doesn't have technical debt.
It has Criteria Debt.
The bottleneck is no longer the code. It's deciding well when AI writes it by default.
A 33-page document with a 4-dimension framing, a 10-minute self-diagnosis, and a decision canvas you can bring to your next 1:1.
Free. Email for the PDF plus weekly newsletter.
Generating code is cheap. Deciding well is still expensive.
For decades the dominant proxy for software productivity has been output speed: lines, commits, story points, PRs closed. AI accelerates exactly that more than any tool before it.
But an engineering team doesn't scale by how much code it produces. It scales by its ability to change direction without breaking the system, without losing context, and without concentrating all the judgment in two or three people.
The problem isn't that AI writes bad code. The problem is accepting changes that nobody fully understands. Local speed can make the global cost worse when review, decision, context, and ownership don't evolve at the same pace as generation.
Four dimensions where Criteria Debt accumulates.
It doesn't show up in Jira as a task or in DORA metrics as a dip. It shows up in the quality of the conversation: how things get reviewed, how things get decided, what gets assumed, and who is left holding the change when something breaks.
Review
Whether what we review are decisions or only code. Whether review is a risk-management mechanism or a style filter.
Decision
Whether the relevant technical decisions are discussed before the code is already written. Whether there are explicit criteria about reversibility.
Context
Whether the code leaves enough information for another team to change it without spending six weeks reconstructing the reasoning.
Ownership
Whether every meaningful change has a clear person or team responsible for maintaining it when it fails, changes, or grows.
When generating code becomes cheap, a new question appears: by what criteria are we accepting what we're changing, for the right reasons, and with clear ownership?
That tension is what I call Criteria Debt.
Three tools you can put to work this week.
- 1
A 4-dimension framing you can bring to your next 1:1
A shared language for naming where the real friction in your team is concentrated.
- 2
A self-diagnosis for your team (10 min)
16 questions, 4 dimensions. Works best with two or four people in parallel: the divergences are the useful part.
- 3
A Technical Decision Canvas you can print this week
One A4 page to raise the conversation before writing, not after the PR.
Download the full document.
It's free. I email it to you and add you to my newsletter, where I write about this every week.
Want to work on this with your team?
Beyond the guide, there's a deeper format: a tailored workshop that takes it from individual reading to a team agreement.
Criteria Debt: from the framing to a team agreement.
A hands-on session so your team walks out with two or three written review and decision agreements, usable starting the following Monday, with owners and a 30-day review date.
It is not an inspirational talk about AI. It is team work on real PRs, recent decisions, and parts of the system where ownership has gone fuzzy.
- Format
- 1 full day or 2 half-days
- Participants
- 8 to 20 people
- Mode
- In person or remote
- Language
- Spanish or English
Who it's for
CTOs, Heads of Engineering, EMs, and Staff+ Engineers whose teams already have AI in production and want to agree on how to review, decide, and share ownership without piling up invisible debt.
Deliverables
- A checklist for AI-assisted PRs adapted to your highest-risk changes.
- A Technical Decision Canvas applied to a real team decision.
- One concrete review or decision rule, written by the team, with an owner and a 30-day review date.
Emilio Carrión
Staff Engineer at Mercadona Tech · PhD candidate at UPV · Software production methods
I'm a Staff Engineer at Mercadona Tech: I build systems at real scale in physical retail. I combine this with a PhD at UPV on software production methods. Outside of work I write for thousands of senior engineers, give talks, and selectively collaborate with technical leaders and Product Engineers who are redefining their craft now that AI writes the code.
I facilitate the workshop outside of my work at Mercadona Tech, with a limited schedule to keep the quality.
I've spoken at
Common questions.
▸Is it AI theory or applied work?
Applied work. We leave with written review and decision agreements, with an owner and a 30-day review date. If you want an inspirational talk about AI, this is not the format.
▸What team size fits?
Between 8 and 20 participants. Above 20 the conversation dilutes; below 8 there aren't enough voices for the useful disagreements to surface.
▸What seniority is it aimed at?
CTOs, Heads of Engineering, EMs, and Staff+ Engineers whose teams already have AI in production. Mixing profiles is welcome: the divergences in judgment between roles are the useful part.
▸Remote or in person?
Both. In person works better for walking out the same day with a written agreement; remote requires splitting the block into two half-days with async work between sessions.
▸What language?
Spanish by default. English available if the team is international.
▸How does it fit with your work at Mercadona Tech?
This is selective work outside of my role at Mercadona Tech. I keep a limited schedule of engagements so as not to compromise quality or hours.
▸What happens after the workshop?
You keep the written agreements, the owners, and the 30-day review date. If you want, that day I run a light review: which agreements held up against reality and which ones need to be rewritten.
Want to bring it to your team?
Write to me with three things: your team size and role, where the friction shows up with a concrete example, and what you've already tried. I reply within 48 working hours, honestly, whatever the answer is.
Write to me: hola@emiliocarrion.com →