What’s possible
Imagined operators, aged 15 to 84, and what Arcadia would take off their hands while their coding agents do the building. None of them are customers.
Arcadia is real software in active development. It is not finished, it has no customers or measured productivity outcomes yet, and early users help decide what it becomes.
Retired shop teacher · Duluth, Minnesota
Marisol describes the tool once. Arcadia keeps that description current, splits it into small steps with plain finish lines, and only calls Marisol back when a choice genuinely belongs to Marisol.
Guide: Coding agent session management with Git worktrees →Emergency-room nurse · Houston, Texas
Between twelve-hour shifts, Kofi answers one decision at a time from a phone. Everything else keeps moving, and nothing reaches colleagues until Kofi says so.
Guide: Governed autonomous development for AI coding agents →Robotics team captain · Pittsburgh, Pennsylvania
The team runs three agent sessions at once. Each gets its own branch and its own action, so nobody’s work overwrites anyone else’s the night before the event.
Guide: Coding agent session management with Git worktrees →Dairy farmer · County Cork, Ireland
Work happens after evening milking. Arcadia remembers exactly where it left off, so Declan never re-explains the herd, the vet’s format, or last week’s half-finished export.
Guide: Coding agent session management with Git worktrees →Pediatric oncologist · Boston, Massachusetts
Every step carries acceptance criteria agreed before work begins. Nothing is marked done because an agent said so, only when the evidence meets what Priya set.
Guide: Governed autonomous development for AI coding agents →Street-food vendor · Lisbon, Portugal
Tomás opens the phone, sees the one next step and why it is next, approves it, and goes back to the grill.
Guide: How to manage multiple AI coding agents →Retired librarian · Asheville, North Carolina
Evelyn is not an engineer and has no wish to become one. “I want people to find Ruth’s stories” becomes a plan with steps and checks, without Evelyn writing the plan.
Guide: AI development for non-engineers: where Arcadia fits →Logistics manager · Newark, New Jersey
When a nightly check fails, the fix is attempted, re-validated, and retried a bounded number of times before Farhan hears about it. Only what needs judgment reaches the morning list.
Guide: Governed autonomous development for AI coding agents →Marine biologist · Tromsø, Norway
Three weeks at sea on patchy satellite. Approved work keeps moving under the gates Ingrid set; anything that needs a decision waits in one list for the next port.
Guide: Governed autonomous development for AI coding agents →Music producer · Atlanta, Georgia
Taking payments is a hard stop. Everything up to the checkout gets built and checked, then the work waits for Jamal’s explicit yes before anything touches money.
Guide: Governed autonomous development for AI coding agents →Elementary-school principal · San Antonio, Texas
Spanish first, no logins: Rosa decides once, and the decision is recorded. No later session quietly builds something Rosa already ruled out.
Guide: Governed autonomous development for AI coding agents →Indie game developer · Seattle, Washington
The dream game is enormous. Arcadia keeps dividing it until the next piece fits in one evening, then rolls each finished slice back up into visible progress.
Linguistics doctoral candidate · Leipzig, Germany
Each analysis script is an action whose criteria name the expected outputs. When the committee asks how Hannah knows it works, the evidence is already on record.
Guide: Governed autonomous development for AI coding agents →Community-college student · Los Angeles, California
Between classes and a night job, Mateo has scattered hours. One queue answers “what’s next?” every time Mateo sits down, with no digging through old chats.
Guide: How to manage multiple AI coding agents →Choir director · Baltimore, Maryland
Messaging members is a gate that is never crossed automatically. Reminders go out only after Grace approves the wording.
Guide: Governed autonomous development for AI coding agents →Retired ship’s engineer · Gothenburg, Sweden
Engine telemetry is a someday idea. It waits with a named trigger — when the new sensor is fitted — so it revives on its own instead of nagging every session.
Microfinance officer · Paris, France
Anything touching member data pauses as a decision, with the consequence of each option spelled out, so Amara can answer in a minute instead of a meeting.
Guide: Governed autonomous development for AI coding agents →Trial lawyer · Chicago, Illinois
Ben bills in six-minute increments. Every handoff ends in a choice with its consequences stated, never a wall of status prose to decode.
Guide: Governed autonomous development for AI coding agents →Rugby coach · Nukuʻalofa, Tonga
Fixtures, fitness, and fundraising all run at once. Each session knows which project it is in and which action it owns.
Guide: Coding agent session management with Git worktrees →Carnatic vocal teacher · Kochi, India
Meera describes the lessons in words. Each finished piece is checked against that intent — the pitch, the rhythm cycle — before it counts as done.
Guide: Governed autonomous development for AI coding agents →Stay-at-home parent · Leeds, England
Oliver gets forty unpredictable minutes at a time. The work pointer means every nap starts exactly where the last one ended.
Guide: Coding agent session management with Git worktrees →Architect · Barcelona, Spain
When a reviewer flags a problem, the fix is made and re-checked, up to a limit, before Yasmin looks. Yasmin reviews finished work, not busywork.
Guide: Governed autonomous development for AI coding agents →Vineyard owner · Sonoma, California
During harvest Diego has no time at all. Approved work continues under standing gates, and the few real decisions queue up for Sunday.
Guide: Governed autonomous development for AI coding agents →Nursing student · Montréal, Québec
Each deck is an action with criteria like “covers every week-four objective.” Done means checked, not claimed.
Guide: Governed autonomous development for AI coding agents →Manufacturing CFO · Detroit, Michigan
Agents work across three repositories. One queue says what matters next across all of them, derived from the priorities Rajesh wrote down, not from whichever tab is open.
Guide: How to manage multiple AI coding agents →Violin maker · Cremona, Italy
Nadia’s one rule — never lose a reading — is written into the project’s constitution. Every agent session receives it before touching anything.
Guide: Governed autonomous development for AI coding agents →City councillor · Kumasi, Ghana
Publishing is an approval boundary. Kwame sees the finished page and the evidence it works, and nothing goes public until Kwame says so.
Guide: Governed autonomous development for AI coding agents →Flamenco dancer · Seville, Spain
Two months on tour, one phone. Lucía answers the questions only Lucía can answer from a hotel lobby; the rest keeps moving without a second thought.
Guide: Governed autonomous development for AI coding agents →Aspiring astronomer · Flagstaff, Arizona
The club learns by reading the plan. Every step has a reason and a check, so they can watch how software gets from an idea to working.
Weaving cooperative leader · Fès, Morocco
Each weaver’s feedback becomes an action, not a lost message, and one queue makes sure the most-requested fix is the next one built.
Guide: How to manage multiple AI coding agents →Orchestra conductor · Copenhagen, Denmark
Should it work offline in the concert hall? The question is asked once, with both consequences stated. Henrik’s answer is recorded and never relitigated.
Guide: Governed autonomous development for AI coding agents →Community organizer · Rio de Janeiro, Brazil
Before anyone depends on it, Ana needs to know it works. Completion means evidence against criteria set in advance, not an AI’s summary of itself.
Guide: Governed autonomous development for AI coding agents →Agronomist · Kisumu, Kenya
The advice rules and their tests get built and verified while Samuel is in the field. Actually texting farmers stays gated on Samuel’s approval.
Guide: Governed autonomous development for AI coding agents →Bakery owner · Lyon, France
Claire wakes at three and wants one answer: is it done? Each finished action comes with its proof, and the one open question sits at the top.
Guide: Governed autonomous development for AI coding agents →First-time founder · Austin, Texas
Five agent sessions and too much ambition. Each session runs in its own isolated worktree on its own action, so speed never turns into collisions.
Guide: Coding agent session management with Git worktrees →Retired nurse and grandmother · Galway, Ireland
Margaret tells a story; the story becomes an action with a clear finish. Any grandchild can pick up the next session without asking what Gran meant.
Guide: AI development for non-engineers: where Arcadia fits →Taxi driver learning to build · Birmingham, England
“Build an app” is overwhelming. Arcadia turns it into a first action small enough to finish before the next pickup, then the next one after that.
Guide: AI development for non-engineers: where Arcadia fits →Museum curator · Naples, Italy
Accuracy over flourish: the curatorial rule lives in one place that every agent reads before writing a single line.
Guide: Governed autonomous development for AI coding agents →Tea-shop franchise owner · Taichung, Taiwan
When the nightly build breaks, it gets repaired and re-validated before opening. Yu-ting hears about it only if it hit a real blocker.
Guide: Governed autonomous development for AI coding agents →Volunteer firefighter · Bozeman, Montana
Every inspection feature ships with acceptance criteria the chief signed off on first. Done is observable, not asserted.
Guide: Governed autonomous development for AI coding agents →User researcher · Toronto, Ontario
Scope keeps growing. Speculative features become deferrals with triggers — “when a second team asks” — instead of silent scope creep.
Sauna builder · Tampere, Finland
Mikael answers questions in the evening, from the sauna bench. Each one is a short list of options with the consequence of every choice, so answering takes a minute.
Guide: Governed autonomous development for AI coding agents →Climate organizer · Johannesburg, South Africa
Twelve volunteers, four coding agents, one source of truth. Every session knows which action it owns and never steps on another’s branch.
Guide: Coding agent session management with Git worktrees →Cardiologist · Bordeaux, France
Medical accuracy is a constraint every session inherits. Pierre reviews finished pages with their evidence attached, never drafts claiming to be done.
Guide: Governed autonomous development for AI coding agents →Surf instructor · Hilo, Hawaiʻi
The day belongs to the ocean. Approved work continues while Keanu is in the water, and pauses only for the choices that need Keanu.
Guide: Governed autonomous development for AI coding agents →Chemistry teacher · Kraków, Poland
What the quiz must check is written as acceptance criteria before anything is built. The agent never grades its own homework.
Guide: Governed autonomous development for AI coding agents →Head chef · Portland, Oregon
Rafael has ideas at one in the morning. Each lands in the queue, ranked against everything else, instead of derailing whatever was in progress.
Guide: How to manage multiple AI coding agents →Coding scholarship student · Abuja, Nigeria
Aisha’s mentor reviews the plan, not every commit. The plan is the thing worth reading: actions, their order, and how each one will be proved.
Guide: Coding agent session management with Git worktrees →Olive grower · Kalamata, Greece
Forty years is daunting. The work starts with one notebook and one verified import, and grows a decade at a time from there.
Product manager · San Francisco, California
Each repository has its own plan and one pointer, so standup is about what is current, not a reconstruction of what happened.
Guide: Coding agent session management with Git worktrees →Cantor · Jerusalem
Moshe decides what the students hear. Any change to the source recordings is a decision that waits for Moshe; everything around it keeps moving.
Guide: Governed autonomous development for AI coding agents →Veterinarian · Uppsala, Sweden
The practice’s messaging keys are credentials, and credentials are never used on their own. Emma runs that one step with a single tap.
Guide: Governed autonomous development for AI coding agents →Bicycle courier · Madrid, Spain
Crowd-sourced data arrives messy. Each import has criteria — bad rows rejected with the field named — so the map stays worth trusting.
Guide: Governed autonomous development for AI coding agents →Stage manager · New York City
Opening night does not move. The vital few features are named and built first, so if time runs out, what gets cut is chosen, not accidental.
Guide: How to manage multiple AI coding agents →Retired railway engineer · Nagoya, Japan
Hiroshi’s tolerances are recorded once as the project’s standing constraints, and every agent session receives them before it starts.
Guide: Governed autonomous development for AI coding agents →Resettlement caseworker · Minneapolis, Minnesota
Personal data is a boundary. The form is built and proved with test data, then it waits for Mariam before anything real touches it.
Guide: Governed autonomous development for AI coding agents →Esports team manager · Milan, Italy
When the calendar integration breaks at two in the morning, the fix runs against the failing check and re-validates. The team wakes up to green.
Guide: Governed autonomous development for AI coding agents →Midwife · Durban, South Africa
“Works offline” is written as an acceptance criterion with a test, not a hope. Thandiwe sees the evidence before the tool reaches a clinic.
Guide: Governed autonomous development for AI coding agents →Whisky distiller · Islay, Scotland
Callum thinks in decades. The intent stays written down, so a session five years from now still understands the cask numbering.
Guide: Coding agent session management with Git worktrees →Data journalist · Kolkata, India
Charts, copy, and data feeds run as separate sessions in parallel, each isolated, each with its own criteria, none waiting on the others.
Guide: Coding agent session management with Git worktrees →Pharmacist · Dearborn, Michigan
Text message or WhatsApp? It is a real trade-off, so it becomes a decision with both consequences stated, and other work moves on while Ahmed thinks.
Guide: Governed autonomous development for AI coding agents →Beekeeper · Freiburg, Germany
Swarm alerts are for later, once the sensors prove reliable. That trigger is written down, so the idea waits without being forgotten.
Lakota language teacher · Rapid City, South Dakota
Every word list is gated on the elders’ review. Nothing is published until that approval is given, and the approval itself is recorded.
Guide: Governed autonomous development for AI coding agents →Fashion student · Florence, Italy
Valentina does not write code. “It has to grade plus sizes correctly” becomes actions with criteria an agent can meet and Valentina can check.
Guide: AI development for non-engineers: where Arcadia fits →Solar installer · Enugu, Nigeria
Crews report problems from the roof. Each defect is captured once, promoted into the queue when it matters, and closed when the fix merges.
Guide: How to manage multiple AI coding agents →Family mediator · Philadelphia, Pennsylvania
Neutrality is a standing constraint: no feature may favour one parent. Every agent session starts from that rule.
Guide: Governed autonomous development for AI coding agents →Avalanche forecaster · Lyngen, Norway
The bulletin pipeline must never be wrong quietly. A defect that blocks it jumps to the top of the queue and stays there until it is fixed.
Guide: Governed autonomous development for AI coding agents →Kindergarten teacher · Kingston, Jamaica
Nia jots ideas on the walk home. Each one-line wish becomes a small action with a clear finish, placed in the queue where it belongs.
Guide: How to manage multiple AI coding agents →Operating-theatre scheduler · Delhi, India
Rohan needs to see why each item is next. The queue shows the priority each action came from, not just the order.
Guide: How to manage multiple AI coding agents →Wildlife photographer · Grenoble, France
Protecting nests is a rule every session inherits. Any feature that might reveal a location stops and waits for Élodie.
Guide: Governed autonomous development for AI coding agents →Cricket club treasurer · Bradford, England
Money stays gated. The scorecards and kit sizes keep moving while Hamza is at the crease.
Guide: Governed autonomous development for AI coding agents →Tai chi instructor · Vancouver, British Columbia
Lin’s students try every version. Their feedback becomes criteria, and an action is done only when those criteria are met.
Guide: Governed autonomous development for AI coding agents →Fintech engineer · London, England
Adaeze sets the gates on Friday night. Approved work runs in isolated sessions; Monday brings one list of decisions and a stack of verified results.
Guide: Governed autonomous development for AI coding agents →Fisherman · The Azores, Portugal
Pedro hates saying anything twice. The project’s purpose and rules live in one document every agent reads, so no session starts from zero.
Guide: Coding agent session management with Git worktrees →Manga artist · Tokyo, Japan
When the deadline looms, the few features that matter this week are named and built first. The rest is deferred, with a reason.
Guide: How to manage multiple AI coding agents →Retired air-traffic controller · Brussels, Belgium
Gabriel trusts procedures, and this is one: clarify, build, validate, record. Every action, every time, with the record to prove it.
Guide: Governed autonomous development for AI coding agents →Special-education teacher · Manchester, England
Sana’s requirements are specific and personal. Recorded once, they shape every action, and nothing reaches a classroom without Sana’s approval.
Guide: Governed autonomous development for AI coding agents →Hospital IT director · Malmö, Sweden
Agents never touch production. That line is a constitutional rule, while tests and refactors happen safely in isolated worktrees.
Guide: Governed autonomous development for AI coding agents →Marathon runner · Iten, Kenya
Mercy trains twice a day and checks in once. The check-in is a short list of choices, each with its consequence; everything else is already handled.
Guide: Governed autonomous development for AI coding agents →Craft brewer · Plzeň, Czechia
The scaling math carries acceptance criteria with worked examples. When an agent says it is right, the examples prove it.
Guide: Governed autonomous development for AI coding agents →Notary · Turin, Italy
Chiara’s deadlines are legal ones. A broken reminder is a stop-the-line defect: first in the queue until it is fixed, before anything else continues.
Guide: Governed autonomous development for AI coding agents →Dance choreographer · Seoul, South Korea
A rehearsal app, a portfolio, and a class-booking site: each is its own project with its own plan, all visible from one place.
Guide: Coding agent session management with Git worktrees →Coffee exporter · Addis Ababa, Ethiopia
Buyers ask for proof. Every traceability feature is verified against its criteria before it is called done, so the proof Ruth shows is real.
Guide: Governed autonomous development for AI coding agents →Gap-year volunteer · Dunedin, New Zealand
Liam’s first real software project: a clear outcome, small actions, stated criteria, and no one pretending unfinished work is finished.
Trekking-tour operator · Cusco, Peru
Days in the mountains, no signal. Routine failures are fixed and re-checked on their own; real blockers wait with a plain note of exactly what is needed.
Guide: Governed autonomous development for AI coding agents →Calligrapher · Istanbul, Türkiye
Omar approves each script style before students ever see it. Publishing is a gate; perfecting is not.
Guide: Governed autonomous development for AI coding agents →Hospice chaplain · Cardiff, Wales
Tone matters more than features. Charlotte’s standards are constraints every session inherits, and any message to a family waits for approval.
Guide: Governed autonomous development for AI coding agents →Mountain guide · Kathmandu, Nepal
When a bar of signal appears at base camp, Tenzin clears the pending decisions from a phone in a minute, each with its consequence stated.
Guide: Governed autonomous development for AI coding agents →Public defender · Oakland, California
Beatriz has no time to manage agents. Arcadia decides what they work on next and checks what they hand back; Beatriz sees only what needs judgment.
Guide: How to manage multiple AI coding agents →High-school football coach · Tulsa, Oklahoma
Friday night’s notes become Monday’s plan, ordered so the one thing the staff needs most — tagged film — lands first.
Guide: How to manage multiple AI coding agents →Computer-science student · Hamburg, Germany
Laila moves between Claude Code, Codex, and OpenCode. The plan, the pointer, and the evidence stay put whichever agent does the work.
Guide: AI coding agent orchestration and governance →Construction foreman · Phoenix, Arizona
A missing hazard is not a typo. Every briefing template carries criteria the safety officer approved before it was built.
Guide: Governed autonomous development for AI coding agents →Bonsai nursery owner · Takamatsu, Japan
Ayumi works slowly and on purpose. Every deferred idea carries the condition that revives it, so nothing is rushed and nothing is forgotten.
Minibus fleet owner · Nairobi, Kenya
Drivers report problems by phone. Each becomes one tracked defect, never a duplicate, fixed in order of what actually blocks the fleet.
Guide: How to manage multiple AI coding agents →Retired judge · Zagreb, Croatia
The guide must never overstate anything. Honesty is a standing constraint, and every page is checked against it before it counts as done.
Guide: Governed autonomous development for AI coding agents →Delivery rider · Amsterdam, Netherlands
Mohammed knows exactly what riders need, but not how to specify it. That knowledge becomes a plan with acceptance criteria — the part that was missing.
Guide: AI development for non-engineers: where Arcadia fits →Children’s author · Stockholm, Sweden
Illustration, narration, and page turns are built as separate actions in parallel sessions, then combined, without any one overwriting another.
Guide: Coding agent session management with Git worktrees →Mechanic · Marseille, France
Samir fixes cars, not software. When a check fails overnight, the repair happens before the garage opens, with a short note of what changed.
Guide: Governed autonomous development for AI coding agents →Retired seamstress · Albuquerque, New Mexico
Lola has never opened a terminal and never needs to. One plain choice a day, with its consequence stated, and the calculator gets better every week.
Guide: AI development for non-engineers: where Arcadia fits →Disaster-response coordinator · Sendai, Japan
Every feature is proved against criteria written in calm times, so nobody discovers a gap in the middle of a crisis.
Guide: Governed autonomous development for AI coding agents →