Squad composition
Six specialist roles collapse into one.
A squad at Level 1 has six job titles and three handoffs. A squad at Level 5 has one title. Not because the work disappeared, but because the coordination did.
L3 · Today, at best
8 dedicated people
Connected human-agent team
- Product manager
- Product owner
- Product designer
- Engineering manager
- 2 Frontend developers
- 2 Backend developers
Agents connect the delivery tasks. The team proves the capacity gain before anyone touches the staffing model.
3 shared specialists
- 1 QA
- 1 Product marketing manager
- 1 Technical writer
L4 · After the 30 days
6 dedicated people
The transition
- Product lead
- Product designer
- 4 Product engineers
Product and delivery roles converge once the autonomous workflow clears the agreed quality and reliability thresholds.
Shared expertise, outside the core team
- QA expertise
- Product marketing
- Documentation expertise
- Engineering leadership and platform
L5 · Where it goes next
~4 product engineers
Target composition
- 4 Product engineers
A small team owns the complete product loop. Each engineer can take a feature from customer feedback to a released, documented outcome.
Shared expertise, outside the core team
- Design expertise on demand
- PMM for positioning and major launches
- QA and documentation expertise
- Engineering leadership and platform
Levels 1, 2 and 3 are the same eight people. Tooling changes, staffing does not. The roster only moves once the workflow has proven itself, which is the whole point of measuring first.
What changes between L3 and L4
PM + PO One product lead
One owner sets the goals and resolves priorities. Agents turn signals into ranked opportunities and executable specs, so the second seat was coordinating a handoff that no longer exists.
4 developers 4 product engineers
The same engineering capacity, a much wider mandate. Agents write, test, review and release. The engineer owns the outcome, from the customer signal that started it to the release note that closes it.
Embedded EM Shared engineering leadership
Leadership and platform work serve several squads at once. QA, documentation and marketing expertise stay available for standards and the calls that need a specialist.
Who owns the work at Level 4
Autonomy is not the absence of accountability. It is accountability moved to a smaller number of larger decisions.
- Human ownership
- Four product engineers own feature outcomes across frontend and backend, from feedback to release and docs. A product lead and a designer support prioritization and the novel experience decisions.
- Agent contribution
- Agents execute routine discovery, implementation, testing, release and documentation. People set goals, make the product and architecture calls, and resolve exceptions.
- Docs, assets, communication
- Documentation ships with the release. Agents prepare internal and customer communication, product engineers verify the claims, the shared PMM approves positioning.
The role that absorbs the rest
From Level 4 onward the squad has one dominant job title: product engineer. Every feature has exactly one of them named against it, accountable for the outcome and not for a stage of the outcome.
They pick up the ranked opportunity, decide what the product should do, judge whether the generated design is right, direct the agents that write and test it, hold the quality bar, own the release, and read the usage data a week later to find out whether any of it worked.
This is a bigger job than any of the roles it replaces, and it is the reason the model works. One person holding the full chain needs no handoff, writes no status update, and attends no alignment meeting. That recovered time is most of the gain.
One person, one chain
- 01 Customer insight
- 02 Priority call
- 03 Spec
- 04 Design judgment
- 05 Implementation
- 06 Tests and quality
- 07 Security review
- 08 Release
- 09 Documentation
- 10 Adoption follow-up
Agents do most of the doing. The product engineer owns every one of these outcomes.
Prove the workflow before you change the staffing.
The eight to six proposal is a hypothesis, not a prerequisite. You validate output, quality, reliability, the number of human interventions and total cost first. Only then does the org chart become a fair conversation.
The four-person model belongs to L5 and L6. It is not what we install in 30 days, and we will not price it as though it were.
An autonomous squad is also not an emptied squad. The people the model frees go to the next squad, the one that has been waiting a year for capacity. We change staffing conclusions only once the numbers are in.
Start with where you actually are.
Twelve questions gives you your level, a score per phase, and the three steps holding you back.