§ 02 / The Agentic Sprint
Pilots stall. Productisation ships.
Four weeks. One production agent loop, deployed into your stack with the evaluation harness, observability, and CI integration that make it survive after we leave.
- Duration
- 3–4 weeks
- Investment
- £20–35k flat
§ Why the sprint exists
Most agentic projects die in the gap between demo and production.
The Sprint exists to close that gap with one well-chosen loop, end to end. You finish with code your team owns - not a prototype that needs another quarter.
$ ./sprint-process.sh
How four weeks run.
We don't sell hours. We sell a deployed loop and the rigour around it.
- week 1
Architecture & evaluation harness
We design the agent loop in detail and build the evaluation harness first. The harness is the contract: it defines success and catches drift before users do.
- week 2
Implementation & integration
We build the loop end-to-end inside your codebase. Real data, real auth, real CI. Daily commits. You see progress in your repo, not in a slide deck.
- week 3
Observability & deployment
Tracing, audit logs, cost-per-run dashboards, and a failure-mode register. We deploy to staging and run the harness against production-like data.
- week 4
Production cutover & handover
Production deployment behind a feature flag. Pair-programming sessions with your engineers so they own it after we leave. Two weeks of post-launch support included.
$ ls deliverables/
What you get.
- ·Production agent loop, deployed to your stack
- ·Evaluation harness with regression suite
- ·Observability dashboards (traces, cost, failure modes)
- ·CI integration with budget guardrails
- ·Architecture documentation + runbook
- ·Pair-programming handover with your engineers
- ·Two weeks of post-launch email + Slack support
$ man faq
Frequently asked.
- Do you only build a single loop?
- Yes. The Sprint is opinionated: one loop, fully productised. If you need multiple, that's a sequence of Sprints - we'll plan it after the first one ships.
- Whose stack do you build in?
- Yours. We commit directly into your repo. We don't sell black-box tools or licensed platforms.
- What languages and frameworks do you support?
- TypeScript, Python, and the major AI frameworks (Anthropic SDK, LangGraph, MCP). For everything else, ask - we adapt.
- Should I do an Audit first?
- If you're not 100% certain which loop to build, yes. The Audit's architectural plan often becomes the Sprint's brief. It de-risks the Sprint significantly.
- What about confidentiality?
- Mutual NDA before any code or data is shared. We don't reuse client code or proprietary logic across engagements.
Ready to scope this?
Book a 30-minute consult. We'll come back inside 48 hours with a fit assessment and a calendar slot.