02 How the pair works
Two parties. One standard.
An engineer and an AI, working as peers on the same branch. The AI is fast; the engineer is hands-on, autonomous and accountable. And because the result is ordinary code, that engineer can be ours, your in-house team, or both.
An engineer and an AI build it together, start to finish.
From the first probe of a source to a pipeline running in production, they work as peers on the same branch. You decide with the engineer at the plan, the review and the ship.
01
Probe
Point it at an API or your warehouse tables. Read-only, safe by construction.
AI02
Profile
Structure, types and key candidates worked out automatically. No values kept.
AI03
Plan
A brief and a plan: what is new, what you already have, open questions.
✋ you approve04
Build
Integration, models, tests and docs, written on one branch by the AI and the engineer, running locally first.
AI + engineer05
Check
Automated gates plus independent reviewers, each covering only their own area.
AI06
Review
Independent reviewers and the engineer read every line. The engineer decides.
✋ you decide07
Ship
Merged to your main under your rules, applied to your cloud only after review.
✋ you approve08
Run
Scheduled, monitored and alerting. Run order and freshness gates built in.
AI + alerts
Who does what
Codes, decides, signs off
Not a reviewer on the sidelines. Pulls the same branch into their own editor and writes code alongside the AI, with full autonomy over how it is built. Owns the architecture, reads every line that ships, and is accountable for the outcome, by name. The engineer can be ours, someone on your team, or a contractor you bring in later: anyone who can read the code can take the keyboard.
Builds at machine speed
Scaffolding, models, tests, migrations, documentation written as the code is, and the long tail nobody wants to do by hand.
What the AI never does
These are hard gates, not guidelines. Each needs an explicit human decision.
- Merge its own work. A person reads and approves every change.
- Apply infrastructure. Plans are reviewed by a person before anything is applied.
- Run account-level grants or change who can access what.
- Rotate keys or handle secrets. Credentials stay in your secret store; code holds references only.
- Publish or delete data. Anything irreversible waits for a yes.
How a change lands
The AI drafts
It writes the change fast, inside your repo, on a branch, with tests and documentation alongside the code.
The engineer joins the branch
They pull it into their own editor, change what they want changed, write the hard parts themselves and push. The AI picks up their commits and builds on them. It is one branch with two authors.
Automated checks run
Secret scanning, conventions, contracts, reconciliation and sensitive-data checks, plus narrow independent reviewers that each cover only their own area.
The engineer decides
Reads every line, and either sends it back or approves it. This is where judgement happens.
It merges to your main
Under your repository's rules, with a person's name on the approval.
Why not only one of them
AI alone
Fast and confident, and sometimes wrong in ways that look right. Nobody is accountable for the result.
Humans alone
Careful, and slow and expensive on the repetitive parts that make up most of a data platform.
The pair
The speed of the first, the judgement and accountability of the second. See it play out on the home page examples.