Skip to the content

B Team (working title)

Question 3 of 5

People and AI, building together

You ask: How do people and AI build together? The role of software engineers is changing, and you have few.

Agents write much of the code. People decide what to build, guide the work and check the result. Reviewers are designed to look at every change, and where you have an engineer, they read what matters most. Where you have none, each change says plainly that no human engineer has read it. Nothing goes live without a person saying yes.

The separate reviewer and the gate that needs a person are built and measured. The standard set of five reviewers and the choice of review level are planned.

How it answers

Builder agents write the change on their own branch. A worker holds no write key. Measured (in part)
A separate reviewer checks the work and cannot change it. Measured (pair runs)
Five standard reviewers, one for each kind of check: security, documentation, compliance, done-when and architecture. Planned, not built
You choose a human review level: A checks only, B an engineer reads risky changes, C an engineer reads all. Level A is decided and says: no human engineer has read this code. Decided 2026-10-13; not built
AI writes the checks, the checks run without AI, and a person approves them. Planned, not built
An agent cannot approve its own work, and assistants cannot answer a gate. Measured, with a recorded deviation

B Team · Artefacts

Volunteer shift planner · Artefacts · change 1 of 5

The swap request screen

The five standard reviewers Each looks at one kind of thing, and gives the file and line it rests on.

Securitypassed with a note

The screen sends the other volunteer's phone number to the browser. It is not needed here.

Cites web/shifts/swap.html line 58

Model claude-opus-5-5 · its own test set: not written yet

Documentationpassed

Screen texts match the wording of the volunteer guide. The guide itself is task T5.

Cites docs/guide.md line 7 to 19

Model claude-opus-5-5 · its own test set: not written yet

Compliancepassed with a note

The swap message carries a volunteer's name and phone number. Your data policy covers names; check that it covers phone numbers.

Cites web/shifts/swap.html line 58

Model claude-opus-5-5 · its own test set: not written yet

"Done when"needs your eyes

Item 1 looks met from the code. The reviewer cannot judge whether the screen is easy to use on a phone. Try it.

Cites web/shifts/swap.html line 12 to 40

Model claude-opus-5-5 · its own test set: not written yet

Architecturepassed with a note

Two texts are written into the page instead of the shared text file the other screens use.

Cites web/shifts/swap.html line 31

Model claude-opus-5-5 · its own test set: not written yet

Be careful what this means. The builder is claude-sonnet-5 and all five reviewers are claude-opus-5-5, which is the same company's models. Your data policy allows one provider only, so a reviewer from a different family is not possible yet.

Human review level A

No human engineer has read this code.

The five reviewers and the checks are the only review. If you accept, you accept it on that basis.

Decide

Accept this change

Send back with this reason

acts now: your decision is recorded in your code host

One change with its five reviewers, each naming the file and line it rests on, and the plain line at review level A. Nobody has shown yet that a reviewer catches a real fault.

Reliability on this page

This page is the "checked" leg: AI writes the checks, the checks run without AI, and a person approves (claims 13 and 70). The reviewer is another agent, so it is checked too: nobody has shown yet that it catches a real fault (claim 10, issue #28). Each standard agent is also meant to ship its own test set (claim 69). How we raise reliability

The honest limit

A team with no engineer has nobody human to read the code. The case that breaks the gate: you approve a change nobody on your side can judge.
Nobody has shown that a reviewer catches a real fault (issue #28). All reviewers use one model family.
Whoever holds a project's key (Temporal token) to the engine's workflow service can send a decision under any name (issue #32). Approval is also only as strong as the reviewer's GitLab login (no issue found yet).

For NGOs

No in-house team is needed to start. Volunteers and a part-time contractor can work with it. Today you still need a terminal to install and run it, and accepting a piece of work is a comment in GitLab.

The technical detail

The claims behind this page, with their states as the evidence page records them.

No.	Claim	State	What it serves	Check or issue
10	A separate reviewer checks work before you see it. Nobody has shown that a reviewer catches a real fault.	Measured (pair runs)	Independent review (F13)	reviewed_checks; #28
11	The reviewer cannot change the work.	Measured	Reviewer without write access (F13, S2)	reviewer_write_checks (5 of 5)
12	The reviewer is a different model family from the worker.	Not built	Different family by default (F13)	#82
13	Nothing is published or merged without a person. No self-approval. The recorded deviation: whoever holds a project's key (Temporal token) to the engine's workflow service can send a decision under any name.	Measured, with a recorded deviation	Human approval (F9, S9)	inbox_checks; #32
22	Agents hold only needed tools. A worker has no write key.	Measured (in part)	Least agency (S2, N4)	reader_write_checks
33	Assistants cannot answer a gate.	Built	Gates closed to assistants (F9, S9)	test_tools (5 tools withheld)
39	A project adds checks and approvers that only raise the bar.	Not built	Project file, command runner (F20)	#198, #199
46	Five standard reviewers, one per kind of review: security, documentation, compliance, Done when, architecture.	Not built	Review follows the risk (F22), compliance reviewer (C10)	#107; #299
47	Your team chooses a human review level, shown on every change. Level A, with the line "no human engineer has read this code", is decided.	Not built	Review follows the risk (F22)	Level A decided 2026-10-13. #299
58	Standard agents, instructions and checks come ready-made. A team adjusts them.	Not built	First project type (no number yet)	#318
69	Each standard agent ships a test set. The system, not a person, enforces the gate.	Not built	Test sets (F5)	#115; #316
70	Code checks what code can. AI-written checks run without AI. Each fault becomes a test.	Not built	Checks by code (F29)	#296

The other four questions

Question 1
Build what actually matters

Start from your goal. End with real users. Judge by use.
Read the answer
Question 2
Choose your models, keep your freedom

You choose your models and where they run.
Read the answer
Question 4
Use AI and stay compliant

B Team produces the records your compliance needs as you work.
Read the answer
Question 5
Fit your people and your budget

Work follows your rhythm. Much of it runs in batches.
Read the answer

Tell us you are interested

B Team is not offered to anyone yet. NGOs come first. If you want to hear when that changes, or you would like to be one of our first pilots, write to us.

Write to: [contact address to be decided]

This draft has no form and collects no data. Once there is an address, what happens next is meant to be this: we tell you whether we can work with your organisation yet. We promise no date and no price.

Next: Use AI and stay compliant