Product & engineering
Ship faster, with quality you can measure.
At a glance
Enterprise customers report that exports fail for large accounts.
- Deployed AI
A day or more
Engineers do every step
- Operational AI
Engineers spend their time on review and design
AI runs triage and first fixes; engineers decide what ships
Illustration
What matters
What product and engineering teams are judged on.
Lead time
From idea to production in days, not sprints.
Quality and reliability
Fewer incidents, and faster recovery from the ones that happen.
Engineering focus
Less time on triage, toil and documentation.
Security
Changes checked for risk before they merge.
Walkthrough
The same request, three ways.
The request
TicketCustomer support
BUG-4127: export fails for large accounts
CSV export times out above 200,000 rows.
Three enterprise customers affected.
Autonomy level 1
Deployed AI
A chat assistant. Someone asks, it answers, people do the rest.
- A person types
Why might a CSV export fail for large datasets?
- AI replies
Common causes include timeouts, memory limits and pagination issues. Consider streaming the export or running it as a background job.
- Person
Engineer. Reproduces the bug, finds the cause, writes the fix and the tests, and updates the ticket.
3 steps · one to a person · a day or more
- How long
- A day or more
- Who does the work
- Engineers do every step
- What keeps it safe
- None: general advice, not your code
- What goes to people
- Everything
Autonomy level 2
Specialised AI
It works from your data and tools and prepares the work. A person approves.
- Trigger
BUG-4127 is filed in the issue tracker.
- Tools
logs.search("export", errors, 7 days, redacted) · repo.search("exportCsv") · tests.run("exports")Returned: Timeouts after 30 seconds above 200,000 rows. The export loads every row into memory.
- AI replies
Proposed change: stream rows in batches. Tests added for small, large and very large accounts.
- Person
Engineer. Reviews the change and merges it.
Approved - Action
Released behind a feature flag for the three customers first, with a one-click rollback.
5 steps · one to a person · about an hour of an engineer's time
- How long
- About an hour of an engineer's time
- Who does the work
- AI investigates and drafts; engineers review
- What keeps it safe
- Every change reviewed and tested
- What goes to people
- Every change
Autonomy level 3
Operational AI
It runs the process within set limits, logs every step and hands people the exceptions.
- Trigger
Every new issue and every failed check.
- Rule
Classify, reproduce and propose fixes, with no write access to production.
Logs are redacted before the AI sees them. Nothing merges without passing tests and an engineer's review.
- Action
Duplicates merged, owners assigned, the fix and its tests proposed, documentation updated.
- Person
Exception: goes to a person
Tech lead. The fix changes how customer data is exported, so it goes to the tech lead for a data and security review.
Escalated - Record
[14:22]Proposal, review and release linked to the issue.
- Report
Weekly: lead time, change failure rate, issues by cause, and toil removed.
6 steps · one to a person · engineers spend their time on review and design
- How long
- Engineers spend their time on review and design
- Who does the work
- AI runs triage and first fixes; engineers decide what ships
- What keeps it safe
- Review gate, test gate, feature flags
- What goes to people
- Every merge; security and data changes
Controls
- Owner
- Head of engineering
- Measure
- Lead time and change failure rate
- Limit
- No production access; review required
- Log
- Every proposal linked to its issue
- Fallback
- Issues queue for the team
- Stop
- Turn off AI proposals in one switch
Illustrative: names, figures and timings are invented.
Where to start
Processes that pay back first.
3Run with exceptions
Issue triage and duplicates
Owners assigned automatically.
2Automate with approval
Tests for new changes
Engineers review each one.
2Automate with approval
Code review for risk and style
A second reviewer, never the only one.
3Run with exceptions
Documentation kept current
Updated from merged changes.
3Run with exceptions
Incident summaries
Timelines drafted from the logs.
1Assist
Architecture decisions
AI researches options; engineers choose.
What we build and run
Built in your systems, run with your team.
- Triage and routing across your issue tracker
- Review and test agents in your pipeline
- Documentation that stays true
- Measurement of delivery and quality
What we measure
Numbers you already track.
- Lead time for changes
- Change failure rate
- Time to restore service
- Hours of toil removed
- Test coverage of new code
The Academy
Learn to run it yourselves: the Product & engineering track.
Ship with AI in the workflow and keep quality measurable.
Being built