Development process

A development pipeline led by an AI agent: a task plan with a Gantt chart, a brief and check verdicts on every task, questions for you, decisions and ideas — with the playbook and the roles kept in the workspace itself.

For anyone building software, a site or a product with an AI agent who wants the agent to follow a clear process rather than whatever it remembers from the chat.

These are examples, not the limit: your agent will build any data model your task needs.

What it is for

An agent writes good code while the task is short. Over weeks it forgets what was agreed, checks its own work and bothers you over trifles — or quietly makes decisions that deserved a word with you.

Here the way of working lives in the workspace itself: a playbook and the roles. The lead agent runs the task plan by it, hands work to a developer, a tester and a reviewer, closes a task only on their PASS verdicts and puts its questions to you in one tab. The workspace rules notice when the process is broken — even after the agent changes or starts a new session.

How it looks in colabro

On the sample data the workspace is created with.

Task plan

Tasks: 8
Planned 2
In progress 1
Checking 1
Waiting for you 1
Done 1
Released 1
Dropped 1

№	Title	Status	Size, steps	Block	After	Tests	Review
1	Project skeleton and test run	Released	S	Foundation		PASS	PASS
2	Class timetable	Done	L	Timetable	1 · Project skeleton and test run	PASS	PASS
3	Booking and cancelling a class	Checking	L	Booking	2 · Class timetable	PASS	FAIL
4	Paying for a pass	Waiting for you	M	Payment	3 · Booking and cancelling a class

and 4 more

“Task plan”: statuses, sizes, order and check verdicts.

Gantt

1 · Project skeleton and test runStep 1–2
2 · Class timetableStep 3–7
8 · Timetable export to a calendarStep 8–9
3 · Booking and cancelling a class in progressStep 10–14
5 · Email reminders in progressStep 10–12
4 · Paying for a pass waitingStep 15–17
7 · Front-desk screenStep 15–19
6 · Student accountStep 18–20

“Gantt”: the tasks in steps, each after the ones it waits for; what is done is on the left.

Gate

Being checked

№	Title	Round	Tests	Review	Has a UI
3	Booking and cancelling a class	1	PASS	FAIL	Has a UI

Playbook violations

All verdicts

№	Title	Status	Round	Tests	Review	Commit
1	Project skeleton and test run	Released	1	PASS	PASS	3f1c2a7
2	Class timetable	Done	2	PASS	PASS	9b04e1d
3	Booking and cancelling a class	Checking	1	PASS	FAIL
4	Paying for a pass	Waiting for you

and 4 more

“Gate”: what is being checked and the verdicts of each task.

Questions for me

Waiting for your answer

Code	Question	Options	Recommendation	Task
Q-001	Cancelling less than 12 hours ahead: should the class come off the pass?	A — always take it off. B — take it off, but the front desk can give it back. C — never t…	B: a clear rule, and the person at the desk handles the odd case.	3 · Booking and cancelling a class

Tasks waiting for you

№	Title	What you need to do	Status
4	Paying for a pass	1. Ask the payment provider the studio's till already uses to enable payments on the webs…	Waiting for you

All questions

Code	Question	Status	Answer	Task
Q-003	How many places per class by default?	Answered	12; 8 in the small room.	2 · Class timetable
Q-002	Which payment provider do we use?	Applied	The till's one — we don't want a second contract.	4 · Paying for a pass
Q-001	Cancelling less than 12 hours ahead: should the class come off the pass?	Waiting		3 · Booking and cancelling a class

“Questions for me”: what waits for your answer or your action.

Playbook

1. Start of every session
This playbook is how work is done in this workspace. It lives here, not in the agent's memory: before any work the agent reads it again. 1. get_workspace and get_changes. Tell the person briefly what they changed since…
2. What is where
The workspace is the shared memory of the project. A role never sees your chat with the person: what is not in the records does not exist for it. Specification (spec) — goal, boundaries, acceptance criteria, risks. Task…
3. Roles and blind checking
The person is the product owner. Sets the goal, answers questions, decides what is expensive to undo, does the steps marked “Needs you”. The agent is the lead. Keeps the specification and the plan, writes briefs, calls…
4. Stage 1. Specification
The input is the person's idea in any shape. The aim is not a nice text but criteria you can test. 1. Study what exists: code, conventions, documents. 2. Turn everything unclear into questions and sort them by “When to…

and 7 more

When the sections changed

№	Section	Changed
1	Start of every session	31.10.2026
2	What is where	31.10.2026
3	Roles and blind checking	31.10.2026
4	Stage 1. Specification	31.10.2026

and 7 more

“Playbook”: how the agent works — sections you can edit.

What the agent does

YouLet's build online booking for our yoga studio.
AgentReads the playbook, writes a specification with testable criteria and boundaries, and cuts it into tasks with sizes and order.
Result“Task plan” and the Gantt chart: what comes after what, and what is already done.
YouHow is the class booking going?
AgentShows the Gate: tests PASS, review FAIL — the lead sends the task back to the developer with nothing but the reviewer's list, for a second round.
ResultThe “Gate” tab: verdicts and reports on every task, no retelling in the chat.
YouWhat do you need from me?
AgentCollects its questions in one batch with options and a recommendation, plus the tasks that need your action, step by step.
Result“Questions for me”: answer right in the table; the agent records it as a decision.

Ask your agent to make one

Copy the request and send it to your agent: it creates the workspace and fits it to you.

Create a colabro workspace from the example dev-process, sample data included (if I have already created it with the button, use that one). Read the “Playbook” document and the roles, then run the project strictly by the playbook: you are the lead, I am the product owner. Re-read the playbook at the start of every session; if you can keep project memory or instructions, save this rule and the workspace address there. Delete the sample tasks, ideas, questions, decisions and specification (keep the playbook and the roles) and start with the specification of my project; put anything unclear to me as one batch under “Questions for me”.

For the agent to do it, connect it to colabro: Connect an agent.

Use this example

You can create the workspace with sample data after you sign in; you will come back to this page.

What people usually adjust

The playbook and the roles are ordinary records: edit a section or a role's instructions and the agent works the new way from its next session (the “Changed” date tells it what to re-read).
Fit the task statuses to your process — the status field of the task type with its x-labels and x-colors; the task-gate rule keeps “Done” behind PASS verdicts.
Task size is in steps (1, 2, 3, 5, 8); the Gantt chart is built from it and from “After”. To plan on dates, add a start field and switch the Gantt block to start.
“Block” often becomes a list of values by part of the project: “Backend”, “UI”, “Docs”.
The sample tasks, ideas, questions, decisions and specification are only there to show the idea: delete them when your own project starts. The playbook and the roles are not samples — keep them.