Three weeks. One use case. A working proof of concept in your AWS account, a cost model, and a written go/no-go you can take to leadership. No retainer, no commitment to the next phase.
The pattern is consistent. A team picks a use case, builds a slide deck, circulates it for a quarter, hires a consultancy that builds another slide deck, and then eventually pays to discover whether the thing actually works on Bedrock.
The Sprint compresses that into fifteen working days. By the end, you have either a working proof of concept that you've seen run on your own data, or evidence that this isn't the right tool for the job. Either outcome is useful. Both save quarters.
Each week ends with a written checkpoint and a working demo of what was built. You're never more than five days away from a clear status update.
Two working sessions, then I write the brief. We agree on the single use case, the evaluation criteria, and the data we'll use to test it.
I build in your AWS account, using Bedrock and whatever supporting services the use case needs. You see commits and a working demo by end of week.
Final demo, written recommendation, and a cost model projected to production scale. If it's a go, we scope what production looks like, separately.
Nothing here is a PowerPoint deliverable. Everything is something a technical reviewer or a CFO can interrogate.
Built on Bedrock, deployed via CloudFormation. You own it on day one. No consultant lock-in, no separate hosting environment to migrate later.
Actual PoC costs from CloudWatch telemetry, plus a projection at production volume. Token economics, infra, and the supporting services around Bedrock.
A 2–4 page written recommendation against the criteria we agreed in Week 1, with evidence and a clear path forward for each outcome.
The things we learned that you don't yet have answers for: data quality, model selection, eval strategy, compliance posture. Ranked by impact.
If the answer is "yes, build it," you'll have a clear path, and you can take that build to me or to anyone else. A production proof of concept runs $25,000 to $50,000, and the Sprint feeds straight into it, so nothing is wasted.
If the answer is "don't build it," you've spent $10,000 to avoid a costly mistake. That's the whole point. If the honest answer is "not yet" or "wrong tool," I'll tell you.
If you're not on this list, I'll tell you on the scoping call and point you somewhere better. That's a feature, not a bug.
Three examples. Names withheld where contracts require it; architecture and outcomes are real.
A Bedrock AgentCore solution that lets studio teams answer cross-system operational questions (project status, storage costs, timelines) in minutes instead of days, pulling answers across separate asset, project, and finance systems.
A multi-agent solution on Amazon Bedrock where a supervisor agent routes plain-language questions to four specialist agents covering driver, asset, customer, and route data.
An autonomous Kanban system and a long-term memory system for AI agents, both running on AWS. I write about how they work, and what broke, on the blog, useful reading before we talk.
My time across the three weeks, the engineering work to build the PoC, the written deliverables, and the final review session. AWS infrastructure costs are billed to your account directly, typically under $200 for the Sprint window itself, but the cost model in Week 3 projects that out properly.
Scope is fixed in Week 1 specifically so this doesn't happen. If something blocks progress mid-sprint (data access, an unknown about your environment) we re-cut scope at the Week 2 checkpoint to land on time. The deliverables ship in three weeks; what's inside them may narrow.
You do. Everything ships into your AWS account and your repository. No consultant infrastructure to migrate off later, no licensing to renew. The CloudFormation templates are yours to extend or throw away.
You get a memo explaining why, with evidence, and you've spent three weeks instead of three quarters. That outcome is more useful than a forced "yes". I'd rather tell you upfront than build something that doesn't earn its keep. Several Sprints have ended this way; the customers came back later with a better use case.
Sometimes. It depends on the use case. The Sprint produces a PoC, not a production system, so a separate engagement scope is the honest answer for most. The Week 3 memo includes a rough estimate of what production would take if you want to keep going.
Tell me what you're trying to do, what data you have, and what "good" would look like. I'll come back with a Sprint scope, a different recommendation, or a pointer to someone better suited.
