Every AI development team has: - Engineers who build - Product people who direct - Designers who shape - Maybe QA who tests
Who: - Preserves context across sessions? - Manages the prompt library? - Documents decisions as they happen? - Ensures learning compounds? - Maintains the system of record?
Nobody. The role doesn't exist.
The Missing Function
This isn't a new task bolted onto existing roles. It's a distinct function:
**Context stewardship.** Ensuring that context isn't lost as work progresses. Capturing decisions. Preserving reasoning.
**Knowledge architecture.** Building and maintaining the structures where knowledge lives. Prompt libraries. Decision logs. Learning repositories.
**Process discipline.** Making sure documentation happens. Reviewing for gaps. Maintaining quality.
**Connection creation.** Linking decisions to evidence. Linking prompts to outcomes. Making the knowledge navigable.
This function is critical for AI-native development. And almost no team has it.
Why It's Missing
**Not historically needed.** Traditional development had natural knowledge preservation through code review, PRDs, and slower pace. AI development moves faster and loses context faster.
**Falls between roles.** Not quite engineering. Not quite PM. Not quite knowledge management. Doesn't fit existing org chart boxes.
**Invisible value.** The benefit is in problems that don't happen. Hard to measure. Easy to deprioritize.
**Seems like overhead.** "We need people building, not documenting." Until nobody understands what was built.
What Happens Without It
**Context evaporates.** Decisions made, reasoning forgotten. Systems nobody understands.
**Prompts scatter.** Everyone has their own prompts. No institutional learning. No leverage.
**Learning doesn't compound.** Same mistakes repeated. Same lessons learned again. No improvement.
**Onboarding suffers.** New people can't ramp. Knowledge is in heads, not systems.
**Velocity becomes chaos.** Fast building, no understanding. Technical debt at scale.
What to Do
**Option 1: Create the role.** Hire or designate someone as context steward / knowledge architect. Make it their job.
**Option 2: Distribute the function.** Build the responsibilities into existing workflows. Everyone does part of it. But someone is accountable for the whole.
**Option 3: Automate what you can.** Tools for capturing decisions, managing prompts, linking artifacts. Reduce the manual burden.
Some combination of all three probably makes sense.
The Skills
If you're creating the role, look for:
- Documentation skill: can write clearly and concisely
- Systems thinking: understands how pieces connect
- Process discipline: follows through, maintains consistency
- Technical understanding: can work with engineers
- Judgment: knows what matters, what doesn't
It's a rare combination. But the function is important enough to find or develop it.
Start Somewhere
You probably can't create the role tomorrow. Start smaller:
- Designate someone accountable for prompt library
- Build decision logging into sprint cadence
- Review for context gaps monthly
Make the function visible. Even imperfectly. Then improve from there.
The role is missing from every AI team. Yours could be the exception.