PHEN
← All case studies

// Case Study

Agentic Delivery Pipeline

A lean crew at BD Performing Arts was buried under a runaway stakeholder backlog. I built them an agentic pipeline that absorbs the PM, QA, and Technical Writer work. The humans pick what gets shipped. Quick tickets close 4× faster, every cycle ships with its own docs, and the system gets sharper every time.

Agentic AIProject Ops

// The Challenge

Lean
crew.

Runaway
backlog.

BD Performing Arts is a small operating team inside a large non-profit. The budget can't carry five full-time roles: PM, QA, Technical Writer, Junior Dev, Scrum Master. The work demands every one of them. Staff, parents, board members, and performers all file requests through one internal tool. Each ask carries a real timeline. Tickets piled up faster than the crew could touch them, and the backlog only ever grew.

So I built them an agentic pipeline that absorbs all five. The agent carries the manual work. The humans carry the judgment. The salaries stay in the building, on programs, not on payroll.

// The pipeline

A request lands and the agent carries it from triage to docs. The humans sign off where it counts. Every cycle ships its own user-facing write-up and feeds the agent's memory for the next ask.

01STAKEHOLDERS Staff, parents, board members, and performers all file through one internal door. Every ask becomes a ticket. 02TASKBOARD A custom Laravel board with its own REST API. Tickets, projects, and sprints sit behind endpoints the agent can call. 03TRIAGE The agent reads every new ticket the moment it lands. Fingerprint dedup catches duplicates; tags route to the right project. 04SCOPE The agent sizes the effort, drafts the scope, and writes the user-facing summary the submitter will see on the ticket. 05APPROVAL The user who filed the ticket reviews and approves the scope. Edits lock so nobody rescopes mid-conversation. REVOKE 06SPRINT Approved tickets rank by effort × human override. The crew fills the queue top-down; the agent picks up the rest. 07BUILD OpenClaw writes the code, opens a branch, and ships a pull request. Every change is traceable to its ticket. 08TEST + VALIDATE Playwright walks the real user flows. An AI reviewer reads every diff and blocks lazy commits before merge. FAIL 09DONE The PR merges. A completion summary lands on the ticket with what changed, why, and where to find it. 10DOCS DELIVERED The system writes user-facing docs and emails the link to the submitter. The agent reads the same docs as memory next time.
From stakeholder ask to delivered docs. Solid arrows are the normal flow. Dashed loops are the two ways back: the user revokes the ticket, or tests fail and the ticket bounces to the dev queue. Every completed cycle leaves docs and a summary behind for the agent to read on the next ask.

// Roles, Offloaded

Five jobs the agent quietly absorbed.

PMQATECH WRITERJUNIOR DEVSCRUM MASTERAGENTHUMAN SIGNOFF
// Five roles in. One agent out. Humans sign where it counts.

// The PM

Triage, size, scope.

Before: someone had to read every ask, size it, and write the scope back to the requester. After a few weeks the queue was the project manager.

After: OpenClaw drafts the triage and the scope in seconds. The human reads it in thirty and signs off.

// The QA

Test, validate, sign off.

Before: smoke tests in someone's notebook, regression sweeps after every release, browser checks by hand.

After: Playwright walks the real flows on every PR. An AI reviewer reads every diff and reruns the suite until green.

// The Tech Writer

Docs that ship with the feature.

Before: docs were a permanent TODO. By the time anyone wrote them, the feature had already changed.

After: each shipped ticket writes its own user-facing doc; the submitter gets the link, the agent reads the same docs as memory next time.

// The Junior Dev

Write the code, open the PR.

Before: small tickets sat in the queue because nobody had a junior dev to throw at the easy work. Senior time got eaten by drudge.

After: OpenClaw writes the code and opens the PR. The senior reads the diff in five minutes instead of writing it in five hours.

// The Scrum Master

Sprint cadence, standup, status.

Before: no one running the cadence. Tickets aged out, blockers stayed buried, and the crew burned cycles on coordination.

After: the agent ranks the queue, surfaces blockers, and emails the daily status. The crew gets a stand-up without a stand-up.

THE OUTCOMES

Five full-time roles' worth of work, absorbed. The non-profit budget stays on programs, not on payroll.

5 FT

Roles not hired

Five salaries the org never had to carry. PM, QA, Tech Writer, Junior Dev, Scrum Master, all absorbed by the agent.

Faster delivery

Quick tickets that used to chew on for weeks close in days. Triage, scoping, code, and review run while the crew sleeps.

Tests + reviews,
on every PR.

QA, fully automated

Playwright walks the real user flows on every PR. An AI reviewer reads every diff, catches lazy commits and security smells, and holds the line on main.

Docs that
feed back in.

Self-improving loop

Every shipped ticket writes its own docs and a completion summary. The submitter gets the link in their inbox. The agent reads its own history on the next similar ask, so the next scope is sharper than the last.

// The Payoff

A small non-profit team that ships like a full department, on a non-profit budget.

Stakeholder questions used to put the crew on its back foot. Now they open the board, point at the ranked list, and walk the stakeholder through what shipped, what is queued, and where the humans overrode the agent.

When a ticket lands in Done, the docs land in the submitter's inbox and the completion summary lands on the ticket. The next similar ask gets a sharper scope because the agent reads its own history. Nothing falls off the board.

Same pipeline runs on phenpedron.com. Same taskboard, same agent, same handoff loop. I built it for myself first, tuned a copy for BD Performing Arts, and I am ready to do the same for the next small team that needs to ship like a full department without the payroll.