Writing a Proposal for Work You Will Build With an Agent

You have scoped the job over a forty-minute call and you know you can deliver it in about two weeks. The agency bidding against you quoted six to eight, and the moment you put a two-week date on paper you either get disbelieved or you commit to a date you set before thinking about their approval process.

That is the proposal problem specific to agent-built work. The build is no longer the long pole, and everything downstream of that has to change.

Price the outcome and the date, never the hours

Bill hours and you are telling the client your value scales with your slowness, and then you go and use the tooling that makes your hours collapse. A day rate for a job you finish in nine days instead of thirty is a pay cut you administered to yourself.

Quote a number and a date. "£9,500. Working system in your hands by 12 September." The client can weigh that against what the thing is worth to them, which is the only comparison that has ever mattered. They cannot weigh 140 hours against anything.

A two-week date on its own invites suspicion. Attached to a milestone schedule with named deliverables on named days, it reads as somebody who has done this before. Same date, different credibility, entirely because of the structure around it.

The structure that works

Six sections, in this order.

The problem, in the client's own words. If the operations manager said "we spend Monday mornings untangling double bookings on the phone", write that, not "inefficiencies in the current resource allocation workflow". This section sells harder than anything else in the document, because it is the only proof you were listening.

What will exist at the end, stated as things they can click. Not "booking module" but "a page where staff see every room for the week and click a slot to book it", and "an admin screen where you add rooms without ringing me".

What is explicitly out of scope. "Not included: payroll integration, a mobile app, single sign-on, migration of the historic spreadsheet, and any changes to the existing intranet." Out-of-scope lists feel defensive while you draft them and turn out to be the most professional part of the document the day a disagreement starts.

Then milestones with dates and payments, the price on its own line, and what you need from them.

"What I need from you" is now the most valuable section

Nobody had to write this down when a build took eight weeks: a client taking nine days to decide who may override a booking was invisible, because you were busy anyway. In a two-week project, nine days is the project. The bottleneck has moved from your keyboard to their inbox, and the proposal has to say so.

Name the artefacts, name a person, name the turnaround.

I need three things from you to hold this date: the list of rooms with capacities, the rule for who may override someone else's booking, and one named person who can approve screens within two working days. If any of those arrives more than a week late, the delivery date moves by the same amount. I will tell you on the day it happens rather than at the end.

Two working days is worth insisting on. Say five and you have priced a fortnight of waiting into a two-week project.

Slice vertically

Traditional phasing was horizontal — schema, then API, then interface — and each phase produced nothing anyone outside the team could use, so clients paid against invisible progress and learned to distrust it.

Agents make vertical slicing cheap, and vertical slicing is what makes a compressed timeline believable. Milestone one is not "database and authentication". It is "staff log in, see this week's rooms, and make a booking that saves" — thin, real, running on a URL they can open on their phone in a meeting. Milestone two adds approvals and the confirmation email.

They see something working on day four. That does more for your credibility than any prose about your process.

Say what happens when they change their mind

They will change their mind, and they should, because seeing it running is when people finally understand what they wanted. The clause has to make that welcome and still get you paid.

Anything not listed under "What this includes" is a change request.
Changes are normal and usually improve the product once you can see it
running. For each one I will give you a fixed price and a revised
delivery date before I start work, and nothing proceeds until you have
replied yes in writing. Email is fine.

Permissive in tone, absolute in mechanism. Note that it puts the date on the table beside the money, which is the half people forget.

Put the hardening milestone in the document

The honest part. Agents get you to something that broadly works very fast, and then there is the real job: empty states, error paths, the person who books a room for 30 February, the layout that collapses at 380px, the migration you have run once on clean data.

Name it, or the proposal implies it does not exist and the client reads the last feature milestone as "done". "Milestone 4 — review and hardening. I go through the whole system as an adversarial user, fix what breaks, load your real data, and run it against an actual Monday morning." Real days, real money.

The AI question, without embarrassment

Take a position and hold it. Do not put "built with AI" in the proposal as a feature, because it is not one — the client is buying a booking system, not a methodology. Do not put it in as a disclaimer either, because a disclaimer implies a defect.

If they ask, answer straight, in one sentence, and move on.

Yes, I use AI coding agents heavily. I review, test and take full responsibility for everything that ships, and it is a large part of why the timeline is what it is.

One rule with no exceptions: never claim a process you are not following. If the proposal says hand-written or bespoke, you have signed a document you are in breach of. Disclosure at handover is a different moment with a different purpose, and it is not optional.

When they ask for a discount because "the AI does the work"

This arrives late, usually from the person who signs. It is not really an argument about AI, it is the oldest procurement move there is wearing a new hat. Do not get defensive and do not move the number.

The price is for the working system and the date it lands on, not the hours behind it. If budget is the constraint I can cut scope — drop reporting from milestone three and it is £8,000. I would rather do that than discount the same work.

Cutting scope for less money keeps your rate intact. A discount on identical scope tells them the first number was invented, and every number you quote afterwards gets treated the same way.

The skeleton, filled in

Internal room-booking tool, around forty staff, replacing a shared spreadsheet. The figures are a worked example at a price band I judge this size of job supports for an experienced independent: £9,500, roughly $12,000.

• Milestone 1, day 4 — login, weekly room grid, create a booking. £2,850 on signature.
• Milestone 2, day 8 — approvals workflow and confirmation emails. £2,850 here.
• Milestone 3, day 11 — admin screens and the monthly utilisation report.
• Milestone 4, day 14 — hardening, real-data load, documentation, handover call. £3,800 on handover.

Thirty per cent up front, thirty at milestone two, forty at completion, totalling £9,500. Two pages plus the out-of-scope list and the change clause.

The odd thing about proposals for agent-built work is how much less they are about you. There is little point competing on how the sausage gets made: the speed is table stakes now and the client cannot audit it anyway. What is left is whether you understood their Monday morning, whether you named the decisions they will have to make, and whether the date you wrote down turns out to be true. That last one is the entire reputation.