“Embracing AI is not just an evolution — it’s a reinvention.” — Cesar Gon, CEO of CI&T
That word matters. Reinvention. Not iteration. Not integration. Not enhancement.
Reinvention means something gets left behind.
The Question Teams Keep Asking
Every time I sit down with a team adopting AI at scale, the same pattern plays out.
We are five minutes into a conversation about autonomous agents, specification-driven workflows, and the new speed of software delivery. The engineers in the room are starting to lean in.
Then someone asks:
“How do I integrate this with Jira?”
Or:
“Should requirements follow the INVEST model?”
These are not bad questions. They are natural questions. They come from experienced professionals who have spent years doing the right thing.
But they are the wrong questions for where we are going. And recognizing why they are the wrong questions is the first and most important step.
Agile Was the Right Answer — for 2003
Let’s be clear: Agile was a genuine breakthrough. Scrum, sprints, INVEST user stories, refinement ceremonies — these weren’t bureaucratic inventions. They were precision-engineered solutions to real constraints:
- Developers could only hold so much context at once
- Teams could only estimate so far ahead
- Stories needed to be Small so one person could finish them in a sprint
- Stories needed to be Negotiable because requirements would change mid-cycle
- Stories needed to be Estimable so the team could commit to a velocity
Every one of those principles exists because the bottleneck was human. Human bandwidth. Human coordination. Human attention span.
Agile did not remove the bottleneck. It managed it brilliantly.
The Bottleneck Has Moved
In 2026, an autonomous AI coding agent can execute what used to be a week-long development story in under ten minutes.
Read that again.
The constraint that Agile was built to manage — the finite bandwidth of human developers — is gone. Or rather, it is no longer the binding constraint.
So what is?
Clarity.
The new bottleneck is the quality of your specification. How precisely can you articulate what you want? How unambiguously can you define what “done” looks like? How completely have you resolved the decisions that, in a human workflow, you could defer to a refinement meeting?
An AI agent does not negotiate scope during sprint planning. It executes against what it was given. Vagueness is not deferred — it is amplified. The hallucinations, the partial implementations, the correction loops that waste hours — they almost always trace back to underspecified input.
This is why retrofitting INVEST onto an AI workflow is not just unnecessary. It actively works against you.
INVEST was designed to keep stories deliberately open so humans could negotiate them later. That deliberate openness is precisely what causes AI agents to fail.
A New Paradigm Needs a New Process
This is the paradigm shift most teams miss. They adopt AI tools. They keep their existing process. They wonder why the results are inconsistent.
It is not the tools. It is the process — because the process was built for different constraints.
The emerging pattern across every serious AI-driven development platform is the same:
Precise Specification → Autonomous Execution → Production-Ready Output
There is no sprint. There is no refinement ceremony. There is no story-pointing meeting. The artifact that matters is not a backlog item. It is a specification.
This does not mean process disappears — it means process front-loads. You invest the thinking before execution, not during it. You resolve ambiguity in the spec, not in code review.
That shift — from managing human execution to enabling autonomous execution — is what AI-SDD was designed for. A phased specification lifecycle (Requirements → Design → Tasks) that produces the clarity an agent needs to execute without drift.
Incremental Change vs. Paradigm Change
I want to be honest here. Not every process needs to be torn down overnight.
AI-SDD integrates with Jira, Azure DevOps, and other existing toolchains. Teams and organizations can adopt it incrementally, layering new practices alongside existing ones. INVEST-style requirements can be incorporated if a team’s context demands it.
But — and this matters — teams need to understand what they are adopting and why. If you bolt AI-SDD onto an unchanged Agile mindset without understanding the underlying shift, you will get marginal gains. You will be running a faster horse, not a car.
Sprint velocity is the wrong unit of measurement entirely.
It is a fundamentally different execution model — and capturing that value requires a fundamentally different way of thinking about the work.
Burning the Ships
Cesar Gon described CI&T’s own transformation as burning the ships — the deliberate act of removing the path back to how things were done before, so that the only direction is forward.
Three years of reinvention. Not iteration. Reinvention.
That metaphor is not about recklessness. It is about commitment. It is about recognizing that you cannot fully inhabit the new paradigm while keeping one foot in the old one.
For engineering teams adopting AI at scale: the sprint board is the ship.
Not the discipline. Not the craft. Not the rigor of breaking work down and thinking it through clearly — that rigor matters more than ever. But the ceremonies, the constraints, and the process artifacts that were designed to manage human limitations that no longer apply.
Those are the ships.
Are you ready to burn yours?
I’m Leonardo Sampaio, a Principal Architect and the author of AI-SDD — a methodology for specification-driven development with autonomous AI agents.