10decoders01
Everyone has an opinion, nobody has a shortlist
Forty use-case ideas across six departments, no shared scoring method, and no way to tell an eight-week build from an eighteen-month one.
GenAI Workshops — LiveMost GenAI workshops end with a plan.
Most GenAI workshops end with a plan.
Ours ends with a number you can defend.
The Agent Proof Sprint ends with a working agent on your own data, its accuracy measured against your own documents, and a production cost model at your volume. Fixed fee, credited in full against your pilot — and refunded in full if no prototype ships.
Three days on site · Accuracy measured on your data · Facilitated by the CTO, not a sales engineer
200+
Engineers
37+
Global clients
4
Global offices
ISO
27001 & 9001
Why teams call us in
The pilot isn't the problem. The eighteen months before it are.
By the time most mid-market teams reach us, they've run two or three GenAI experiments that looked promising in a demo and died on contact with real data, real permissions, or real procurement. These are the patterns we see most.
02
The demo worked because the data was cherry-picked
Vendor POCs run on ten clean sample documents. Your actual estate is scanned faxes, three EHR exports and a shared drive nobody has audited since 2019.
03
Compliance found out last
PHI, PII and residency questions surface at security review, not at design. The build stalls while legal reverse-engineers what was already shipped.
04
Nobody can price it
Token cost, retrieval infrastructure, human review and ongoing evaluation never make it into the business case, so the ROI number collapses under scrutiny.
05
The team is one person deep
A single enthusiastic engineer holds all the context. When they take leave or leave outright, the initiative quietly stops.
06
The board wants a number by Friday
Leadership has committed to an AI position publicly, and the engineering answer is still “it depends” — which reads as no answer at all.
What happens in the room
Four formats. One of them is the point.
Each format is self-contained — you can stop after any one of them and still have something usable. Most teams start at 02. If you're not sure, say so on the call and we'll tell you which one fits, including if the answer is none of them.
Workshop 01
AI Literacy & Guardrails
A shared, non-technical vocabulary for the people who approve budget and carry the risk. No model architecture, no prompt tricks — what these systems can and cannot do, where the liability actually sits, and what “we use AI” commits you to in front of a regulator or an enterprise customer's security questionnaire.
What we cover
- What GenAI reliably does, and the failure modes vendors don't demo
- Where your obligations come from — sector rules, contracts, data residency
- Reading an AI vendor claim: the four questions that separate product from wrapper
- Acceptable-use and human-review policy, drafted in the room
What you leave with
- A one-page internal AI acceptable-use position, in your own language
- A literacy baseline across the leadership group, scored
- A named owner and review cadence for AI risk
- A shortlist of what to stop doing immediately
Workshop 02 · Day one of the Proof Sprint
Opportunity Framing Sprint
We map your actual workflows, not a generic industry value chain, and score every candidate use case on the two axes that decide whether it ships: how ready the underlying data is, and how much a wrong answer costs. Most ideas die here. That's the point — killing eight bad use cases in a day is worth more than starting a ninth.
What we cover
- Workflow teardown of two or three high-volume processes you choose
- Data-readiness audit against what each use case would actually need
- Cost-of-error scoring: which decisions can tolerate a machine being wrong
- Build / buy / don't-do calls on each candidate, with reasoning recorded
What you leave with
- A ranked use-case shortlist with effort and data-gap estimates against each
- A written kill list — what you're deliberately not doing, and why
- A first-pass cost model including inference, retrieval and human review
- A sequenced 90-day plan owned by named people on your side
Workshop 03 · Days two and three of the Proof Sprint
Agent Prototype Lab
Two days, one use case, running code. We build on our Rapid Agent Builder framework against a sample of your real data — under NDA, in your environment where possible — so what you see on day two is behaviour on your documents, not a curated demo. You keep the prototype and the evaluation harness whether or not you engage us further.
What we cover
- Scoping the single use case down to something provable in two days
- Ingestion and grounding against your sample data, with citations on every output
- Building the evaluation set — what “good” means, in numbers, before we build
- Where the human stays in the loop, and what happens when the model is unsure
What you leave with
- A working prototype and its source, yours to keep
- An evaluation harness with accuracy measured on your data, not a benchmark
- A production gap list: what's missing between this and something you'd trust
- A go / no-go recommendation with the reasoning written down
Workshop 04 · Sector-specific
Healthcare & BFSI Deep-Dive
The same build discipline as Workshop 03, run inside a regulated workflow — pre-denial claim validation, prior-auth document review, AML alert triage, or sanctions screening. We work against de-identified or synthetic data unless a BAA is in place, and the compliance constraint is a design input from hour one rather than a review gate at the end.
What we cover
- The regulated workflow end to end, including the manual steps nobody documented
- PHI / PII handling, residency and audit-trail requirements as build constraints
- Prototype against de-identified or synthetic data, patterned on your real formats
- What an auditor or payer would need to see to accept a machine-assisted decision
What you leave with
- A working prototype on a regulated workflow, with full decision traceability
- A control map: which requirement each design choice answers
- An explainability pack you can put in front of compliance or an external auditor
- A staged path to production with the review gates already identified
The number that matters
The sprint isn't the expensive part. The wrong build is.
A stalled pilot rarely fails loudly. It absorbs two engineers for nine months, produces something nobody trusts enough to put in front of a customer, and quietly ends. Before you spend that, it's worth knowing what the workflow is actually costing you today.
What is this workflow costing you?
What you keep
Seven artefacts, whether you proceed or not.
Everything below leaves with you at the end of the sprint. There is no clawback and no obligation to engage us afterwards.
Working prototype
A running agent on your own data, source included. Not a demo environment — something you can put in front of the people who'd use it.
Evaluation harness
Your definition of “good” expressed in numbers, with accuracy measured against it on your documents rather than a public benchmark.
Production cost model
Inference, retrieval, human review and ongoing evaluation, projected at your volume. The figure your CFO will ask for and most vendors can't produce.
Workflow teardown
Two or three of your high-volume processes mapped end to end, including the manual steps nobody ever documented.
Ranked shortlist and kill list
Every candidate scored on data readiness and cost-of-error — and the rejections written down with reasons, so the debate doesn't restart next quarter.
Production gap list
Everything standing between this prototype and something you'd trust in front of a customer, sized and sequenced.
Go / no-go memo
A written recommendation with the reasoning attached, signed by the CTO. Board-ready as written, including when the answer is no.
All of it is yours
Written into the engagement terms rather than negotiated at the end — whether or not you work with us afterwards.
Scope your sprint →Engagement formats
Three depths. We'll tell you which one fits on the call.
Every engagement is a fixed fee agreed before we start — no hourly billing, no change orders, no separately invoiced discovery phase. What it costs depends on the depth, the workflow and whether regulated data is in scope, which is a ten-minute conversation rather than a form.
Format 01
Signal
Half day · remote or on site · up to 20 people
- AI literacy and guardrails for leadership
- Acceptable-use position drafted in the room
- Literacy baseline across the leadership group, scored
- Named AI risk owner and review cadence
- Stop-doing shortlist
Right for you ifthe gap is board-level understanding and there's no engineering question on the table yet. The fee credits in full against a Proof Sprint.
Format 02
Agent Proof Sprint
3 days · on site · 6–12 people
- Everything in Signal
- Working prototype on your own data
- Evaluation harness, accuracy measured
- Workflow teardown and ranked shortlist
- Production cost model at your volume
- Production gap list
- Written go / no-go memo, signed by the CTO
- Full source and IP handover
- 30 days post-sprint access
Right for you ifyou want a build answer rather than a strategy answer. The fee is credited in full against a 90-day pilot — proceed, and the sprint costs you nothing.
Format 03
Regulated Proof
5 days over 2 weeks · BAA before day one
- Everything in the Proof Sprint
- Two regulated workflows, not one
- PHI / PII handling under BAA
- Compliance control map
- Explainability pack for auditor or payer
- Staged production path with review gates
- 90-day advisory — two hours weekly
- Named engineer continuity into the pilot
Right for you if the workflow touches PHI or sits under an auditor, and the compliance answer has to hold up alongside the technical one.
Risk reversal
We carry the risk, not you.
Three guarantees, written into the engagement terms. None is conditional on you buying anything afterwards.
01
The Running Code Guarantee
If you do not leave day three with a working prototype operating on your own data, you pay nothing. Not a partial refund, not a credit note — the full fee returned.
02
The Pilot Credit
Start a 90-day pilot with us within 60 days of the sprint and 100% of the fee is credited against it. If you proceed, the sprint cost you nothing.
03
The Honest No
If we conclude you should not build it, we say so in writing and you keep every artefact. We would rather lose the pilot than sell you a build that fails in month four.
There are two outcomes. Either you get a working prototype, measured accuracy and a cost model — and the fee comes back the moment the pilot starts. Or you don't, and you pay nothing.
How we run them
Frame. Ground. Build. Govern.
The same four movements run through every format — only the depth changes. It's the sequence we use on paid engagements, compressed into the time available.
01
Frame
Start from the business problem and the cost of getting it wrong — not from a capability looking for somewhere to land.
02
Ground
Test the idea against the data you actually hold. Most use cases fail here, and finding that out in a day is the cheapest failure available.
03
Build
Put something running in front of the people who'd use it. Opinions about AI converge fast once there's a real output on screen.
04
Govern
Decide the review points, audit trail and escalation path while the design is still cheap to change.
What's different here
Four things we do that the big firms structurally can't.
You keep the code
The prototype, the evaluation harness and the source leave with you, whether you engage us afterwards or not. It's written into the engagement terms rather than negotiated at the end.
No hyperscaler is paying for this
Most vendor workshops are co-funded by a cloud provider, which is why the recommendation always lands on that provider's stack. Ours aren't, so model and platform choice stays an open engineering question.
The person facilitating is the person who'd build it
Thomas runs these sessions himself. There is no handoff from a workshop team to a delivery team, and no incentive to scope something that sounds good and delivers badly.
We'll tell you not to build it
A significant share of use cases that reach us shouldn't be built — the data isn't there, the volume doesn't justify it, or a rules engine does the job for a fraction of the cost. Saying so is part of the deliverable.
Honest comparison
How this sits against the alternatives.
There are good reasons to choose someone else. Here's the shape of the market as we see it, so you can rule us out quickly if we're the wrong fit.
| Option | Typical length | What you walk out with | Choose them when |
|---|---|---|---|
| Global SI workshop programme | 4 hours – 1.5 days | Prioritised opportunity map, maturity score, follow-on proposal | You need a globally consistent programme across many business units and a brand name your board already recognises. |
| Hyperscaler-funded workshop | 2 hours – 1 day | Use-case shortlist scored against that provider's services | You've already standardised on one cloud and want the fastest possible route to a funded POC on it. |
| Boutique executive session | Half day | Leadership alignment, risk framing, discussion guide | The gap is purely board-level understanding and there's no engineering question yet. |
| Internal, run by your own team | Whatever you can spare | Whatever your team already knows, made explicit | You have in-house people who've shipped production GenAI before. If you do, run it yourself — it'll be better and cheaper. |
| 10decoders | 3 days | Running prototype on your data, evaluation harness, cost model, written go / no-go — fee credited against your pilot | You want a build answer rather than a strategy answer, you're mid-market, and you'd rather be told no early than sold a programme. |
Categories describe common market patterns, not any single named provider. Durations reflect publicly published offerings as of August 2026.
Qualification
Worth your three days, or not.
This works well when
- You're a mid-market healthcare, fintech or BFSI organisation with a real operational bottleneck
- Someone with budget authority will be in the room for at least part of the session
- You can share sample data under NDA, even a small or de-identified slice
- You've had at least one GenAI attempt that didn't land and want to know why
- You want a defensible internal recommendation more than you want a vendor pitch
Don't book this if
- You need a signed statement of work this quarter — a workshop will slow that down, so let's just scope directly
- Nobody can clear three consecutive days without a laptop open
- The decision has already been made and you need external validation for it
- You have an experienced in-house GenAI team — you'll get more from running this internally
- No sample data can be shared in any form, including synthetic — we'd be guessing alongside you
Availability
Why this is capacity-limited, honestly.
The differentiator is that our CTO facilitates every sprint personally, and the engineers in the room are the ones who would do the build. That is also the constraint.
3
Proof Sprints per quarter
Any more and either the sprint or the delivery work behind it degrades. We would rather turn one away than run four badly.
6
Founding-cohort engagements
The first six book at the rates on this page. Rates are reviewed after that, and the founding cohort keeps theirs.
31 Dec
Rates held to this date
Anything booked and scheduled before 31 December 2026 is honoured at these figures, whenever it actually runs.
Regulated Proof requires a BAA in place before day one — typically two to three weeks. Start that conversation early if the workflow touches PHI.
Practical questions
Before you book.
Why isn't the price on this page?
Because the honest answer depends on three things we can establish in ten minutes: which format fits, how many workflows are in scope, and whether regulated data is involved. Publishing one figure would mean either overcharging the simple engagements or underscoping the complex ones. What we will commit to in writing before you decide: it is a fixed fee, agreed up front, with no hourly billing, no change orders and no separately invoiced discovery phase. You get the number on the first call, not after a proposal process.
What exactly triggers the refund?
One condition: no working prototype operating on your data by the end of day three. If that happens the full fee is returned — not prorated, not credited. All we need from you is the sample data and the people in the room; if either looks unlikely on your side we will say so before day one rather than after.
How does the pilot credit work?
Start a 90-day pilot with us within 60 days of the sprint and 100% of the sprint fee is credited against it. If you proceed, the sprint effectively cost you nothing. The credit is written into the engagement terms, not offered as a closing concession, and it applies whichever format you ran.
Do you need our real data?
For Workshops 01 and 02, no — those run on your process knowledge and a whiteboard. For 03 and 04 we need a representative sample, which can be de-identified or synthetic as long as it matches the structure, messiness and edge cases of the real thing. Clean sample data produces a prototype that lies to you. An NDA is signed before anything is shared, and for healthcare workflows involving PHI we sign a BAA before access.
Who actually runs the session?
Thomas, our CTO, facilitates. For Workshop 03 and 04 he's joined by one or two engineers from the practice who'd be on the build if it proceeds. We don't staff these with a separate pre-sales function.
Can this run remotely?
Workshops 01 and 02 run well remotely in two shorter blocks rather than one long one. Workshop 03 and 04 are meaningfully better on site — the value comes from engineering and business people reacting to the same screen in real time, and that degrades over video. We'll run them remotely if travel isn't practical, but we'll say up front that you're getting less.
What if we decide not to build anything afterwards?
That's a legitimate outcome and a cheap one. You keep every artefact — prototype, evaluation harness, cost model, written recommendation. There's no clawback and no obligation to proceed. A documented decision not to build, with the reasoning attached, is worth more to most boards than another stalled pilot.
How is this different from your discovery call?
The discovery call is thirty minutes, free, and exists to work out whether there's anything here worth doing. A workshop is a paid, structured engagement with defined deliverables. If a call is enough to answer your question, we'll tell you that rather than sell you a workshop.
How far out are you booking?
We run three Proof Sprints per quarter because the CTO facilitates each one personally. Regulated Proof needs a BAA in place before day one, which typically adds two to three weeks. Ask on the call and we will give you the actual next available window rather than manufacturing urgency.
