Agentic Proof of Concept

A short fixed-price engagement, two to four weeks at Tenhaw, that takes one real workflow and builds a working agentic system against it, in your environment and against your data. The point is to replace an argument about feasibility with a working thing people can use. It is not a production deployment; productionising is scoped and costed separately.

By treating the requirements as the source code. Every requirement is converted into structured markdown, mapped for relationships, and interrogated for gaps and contradictions before any code is written; the build then runs against the whole requirement set at maximum model reasoning rather than file by file. The full method is published at tenhaw.com/the-tenhaw-way/building-with-ai.

The build is pair-programmed with your engineers throughout, deliberately. On a recent engagement the client engineer who paired on a two-week build finished it saying they were 70% confident they could run the process unaided. Seventy per cent after a fortnight is the measured figure, and it is the difference between buying a proof of concept and starting to acquire a capability.

Tenhaw prices agentic proofs of concept between £20,000 and £55,000 as a fixed fee for two to four weeks, depending on the complexity of the workflow and the state of the underlying data. The price is agreed before the work starts and does not move.

No. Tenhaw is not a staffing agency and does not place people by the day into someone else's plan, which is what lets us publish a rate card and stay accountable for the outcome. Senior people are supplied on an engagement with partner oversight behind them. If you want a single senior lead rather than a build, the closest thing on the ladder is programme and delivery management, where one lead runs delivery and governance without Tenhaw building any of it.

You get a documented answer to a question that would otherwise have cost far more to answer, plus the requirement corpus and gap analysis, which retain value regardless. A proof of concept that establishes a workflow is not viable has done its job, and we would rather tell you that in week three than in month nine.

The complexity of the workflow, and the state of the data underneath it. Those set the two variables the price is built from, whether one or two Tenhaw engineers are on it and whether it runs two weeks or four, costed from the published senior practitioner rate of £1,250 a day. A tangled workflow with scattered data needs more of both and sits nearer £55,000; a well understood one with its data already in a single place sits nearer £20,000. Either way the number is fixed in writing before anyone starts, and the deliverable is fixed alongside it.

Access, experts and engineers. The system is built in your environment against your real workflow, so we need that environment and the workflow's data from the first day. Your subject-matter experts are wanted around days four and five, once the gap-and-contradiction pass has surfaced what the written requirements contradict or leave out. And engineers of your own to pair with ours throughout, because your people finishing able to run the method is a measured deliverable rather than a hope. Data is the one hard prerequisite, and where it does not yet exist in any usable form, a proof of concept is the wrong place to start.

Days one to three turn your existing requirements, the PDFs, diagrams and decks, into a structured markdown corpus, map the relationships and run the gap-and-contradiction pass, all before any code exists. Days four and five resolve those gaps with your subject-matter experts and record explicitly what we are proceeding without, and why. Week two is the build, paired with your engineers, with a security review roughly every fifth prompt, and it typically reaches around 80% correct by the end of it. Weeks three and four iterate toward roughly 95%, fold user-testing feedback back through the same corpus, and scope what production readiness would require.

No. It ends on a date agreed at the start with a fixed-price deliverable, nothing rolls on, and there is no retainer waiting on the other side. You keep the working system, the full requirement corpus as structured markdown, and a costed scope for what productionising it would take, all of it yours to extend or to hand to somebody else. Whether to build for production is a separate decision, and it is a better decision for being made with a working thing and a real number in front of you rather than bought in advance.

In three situations, all of them cheaper to spot now. An agentic proof of concept is the wrong buy when you do not yet know which workflow to attack, and the audit is the better answer there at £44,000 fixed over four weeks. It is wrong when what you actually need is a production deployment, because this ends in a proof of concept and hardening one for production is a longer, separately scoped job. And it is wrong where the data underneath the workflow does not yet exist in any usable form, because the fortnight then goes on building the source rather than the system.

No, and the narrowness is the point. One real workflow, in your environment and against your data, is what makes a fixed price over two to four weeks honest rather than a guess, and it is what leaves you with a measured result instead of two half-built things. A second workflow is a second engagement, scoped and priced on its own inside the £20,000 to £55,000 band. If choosing between candidates is the real problem, take the one whose result would settle the argument in the room, because that is where a working system earns the most.

Yes, and that is the normal starting point. Nothing gets built before the requirements are dealt with. Whatever exists, the PDFs, diagrams and decks, is turned into a structured markdown corpus with the relationships mapped, then interrogated for gaps and contradictions. What that surfaces becomes the agenda for the sessions with your subject-matter experts, and anything still unresolved is recorded as something we are proceeding without, and why. So there is no tidying exercise to run before we start. The corpus is the tidying, and it is yours to keep and extend whatever you decide about the build.

They build, in your environment, next to your people. One or two of them are the whole team on this rung, at the published senior practitioner rate of £1,250 a day, each one someone James Rooney has already delivered alongside. The day-to-day work is turning the requirement corpus into a working system against your real data, pairing with your engineers from the first day so the method stays behind, and running a security review roughly every fifth prompt. If what you want instead is a senior person sitting in your stand-up for six months, that is the build team rather than this.

Yes, in weeks three and four. Week two produces the first working build, paired with your engineers, and the fortnight after it folds user-testing feedback back through the same requirement corpus rather than patching the code around it, iterating toward roughly 95% correct. That is also why the engagement ends with something your executives can use rather than a deck about one. It remains a proof of concept, so the testing answers whether the workflow works, and what a production rollout would take is scoped and costed as its own decision.

Four things: a working system your executives can use rather than a deck about one, a measured read on how much of the method your own people absorbed, a costed scope for productionisation as a separate decision, and evidence of whether this workflow is worth pursuing at all. Boards tend to underrate the last one, because evidence that a workflow is not worth pursuing is far more useful before a build is budgeted than after. And if the difficulty is that a plan will not persuade the sceptics in the room, a working system usually does.

Whoever you choose. You finish holding the working system, the requirement corpus and a costed scope for production, so the options are real ones: your own engineers, a different supplier entirely, or a Tenhaw agentic build team at £70,000 to £85,000 a month. Weeks three and four exist partly to produce that scope and that number, so the production question gets answered with a working thing and a real cost on the table. Your engineers having paired with ours throughout is what makes the first option a genuine one rather than a courtesy.

We do. The number is agreed in writing before anyone starts and it does not move, and the deliverable is fixed alongside it, so an overrun is our problem rather than a change request pointed at you. That is a sane promise rather than a gamble because of where the unknowns surface. The requirements are interrogated for gaps and contradictions in week one, before any code exists, and whatever cannot be resolved is written down at the time. Fixed-price work goes wrong when that discovery lands in week three, and this sequence is built to stop it.

Because a free vendor pilot is scoped to prove that product, and Tenhaw has no model, platform or licence to resell or take a margin on. What the fixed £20,000 to £55,000 buys is a working system in your own environment against your own data, one or two engineers pairing with yours while it is built, and the requirement corpus that produced it. If you were always going to buy that product anyway, let the vendor prove it and keep your money. What neither exercise can settle from outside is whether the workflow is worth automating at all, and that turns on volumes and the cost of an error, which only your operation knows.

Not on its own, and Tenhaw scopes that gap rather than glossing it. A proof of concept is built to answer one question, whether this workflow can be done at all, in your environment, against your data, inside two to four weeks. Throughput, error handling and identity carry deliberate shortcuts at that stage, which is why weeks three and four end with a costed assessment of what productionising would take rather than a claim that you are finished. The £20,000 to £55,000 buys the first answer cheaply. What sets the second is your peak volume and how unevenly it arrives, which your operations people already know and no fortnight can discover.

Masked data usually works, synthetic data usually does not, and since Tenhaw builds inside your environment under your own policies, with UK data residency by default and EU available, the choice is yours to make. The test is whether the mask preserves the features the decisions actually turn on. Replace a customer name and nothing changes; flatten the messy free text and the result flatters itself. Synthetic records fail that test because they are generated from the rules you already know, so the exceptions quietly vanish. The one hard prerequisite is data that exists in usable form, and whether your masked copy still carries the awkward cases is a question for your data owners.

In writing, before the fee is fixed. Tenhaw prices a proof of concept at £20,000 to £55,000 against a fixed deliverable, so what counts as working has to be settled first, and the reference it is settled against is the requirement corpus. Your existing documents are turned into structured markdown, interrogated for gaps and contradictions, with anything your subject-matter experts cannot resolve recorded explicitly as something the build proceeds without. Acceptance is then measured against that, rather than against somebody's memory of a meeting. What the corpus cannot decide for you is which specific cases must be right rather than what percentage, because that is a judgement about consequence your process owner owns.

On Tenhaw's own live engagement, a proof of concept built in two weeks is being productionised over four to six weeks with a dedicated team. Treat that as the shape rather than a quote for yours. Weeks three and four of your own engagement exist partly to produce the number for the specific thing you have just watched work, which is why the costed assessment is a deliverable of the fixed fee rather than a follow-on proposal. Three things move it: the integration surface, the non-functional requirements your estate imposes, and how accurate the workflow must be before it runs unchecked. The last is a risk decision rather than an engineering one.

Your own engineers, and Tenhaw shapes the two to four weeks so that is realistic rather than a hope. They pair on the build from the first day, the code sits in your repositories as it is written, and the requirement corpus behind it is structured markdown they can read and extend, so a fix is a change to something they already know rather than a call to us. Be clear-eyed about what you are running, though. It was built to answer a question quickly, so it is a thing to learn from while you decide rather than one to leave unattended in front of customers. How much attention it needs meanwhile depends on who is using it, and how often.

If you have engineers who have already shipped an agentic system and can be freed for a fortnight, run it yourselves, because Tenhaw publishes the method free, failure modes included. What £20,000 to £55,000 buys is that the fortnight actually happens at a fixed price, with one or two engineers at the published senior practitioner rate of £1,250 a day whose only job for two to four weeks is this one, and a gap-and-contradiction pass run by somebody with no stake in which answer it produces. Internal attempts stall on the first of those far more often than the second. Whether your people can genuinely come off their current commitments is the thing to test before you decide.

Yours, wherever you have an approved enterprise tenancy. Tenhaw does not resell or mark up models, platforms or licences, so the build runs inside your infrastructure on your own contracts, and nothing you specify becomes revenue for us. Where there is no approved stack, the default is named up front as Claude Code with OpenAI models and Gemini, best tool for the case, swapped for your preferred stack on request. The £20,000 to £55,000 is the fee for the work, costed from the published £1,250 senior practitioner day rate, with no licence margin underneath it. Whether your existing agreements cover the models this workflow needs is worth checking early, since procuring a new one can outlast a two-week build.

No, and the reason is worth knowing. Tenhaw prices the proof of concept standalone at £20,000 to £55,000, fixed, ending on the date agreed at kickoff. A fee credited against a later build would quietly put a price on the recommendation to stop, and a proof of concept that establishes a workflow is not viable has done its job. What you carry into the production decision instead is a working system, the requirement corpus and a costed scope, so the next number is a real one whoever ends up spending it. Which route that turns out to be depends on engineering capacity you have and we cannot see from here.