AI production sprint
One system in production,
in four to six weeks
For a software company that needs one AI feature in production, an AI production sprint is four to six weeks. We ship one search-and-answer system or one agent, with evaluations, monitoring and a runbook. After that you run it, or Managed AI operations keeps it healthy.
The facts
At a glance.
- Who it is for
- A software company that needs one AI feature in production.
- Length
- Four to six weeks.
- What you get
- One search-and-answer system or one agent, with a golden evaluation set, monitoring, a runbook, and a go or no-go to scale.
- What happens after
- You run it, your team runs it from the runbook, or Managed AI operations keeps it healthy.
- Pricing
- Contact us for pricing.
One workflow
Tell us which system should go live
Internal search and answers, a customer or support agent, or an operations agent. We say whether it fits four to six weeks, and what the fixed price is, before we start.
What we ship
Three variants.
We pick one. The next system is another sprint, or a change order after this one is live.
Internal search and answers
Search and answers over your own documents, in your environment. People can see which source an answer came from.
Customer or support agent
It reads the case, looks up the source, and drafts or takes the next step. It hands the case to a person when the step needs one.
Operations tool-calling agent
It calls the tools you already use, with permissions, a log of each call, and a stop when the action needs a human.
How it works
How the four to six weeks go
Scope and price are fixed before we write the code.
Name the workflow
One system, one boundary, the people who will use it, and how we will know it is good enough. You get that in writing, with the price.
Build it in your environment
Private or virtual private cloud on Amazon Web Services, Microsoft Azure, Google Cloud or GleSYS, or on your own servers, by default. Data stays with you. Audit logs. We do not train on your data unless you ask us to, in writing.
Ship the evaluations with it
A golden evaluation set, a continuous integration gate, monitoring hooks for quality, latency and cost, and a runbook: what it does, who to call, how to turn it off.
Go or no-go to scale
A written recommendation: keep it as it is, extend it under a new fixed price, or stop. Scaling is not assumed.
Scope and price
are fixed before
we write the code
One system in production
Weeks
Evaluations from
the start, not
a later phase
What you leave with
The system, and the proof it still works.
A demo is not the delivery. These ship with the system.
Golden evaluation set
Real examples, agreed with you, that every later change has to pass.
Monitoring hooks
Quality, latency and cost, so someone can see when it gets worse.
Runbook
What it does, which data it touches, who to call, and how to turn it off.
Go or no-go
A short written call on whether to scale, and what that next step would be.
Private model
When the evaluations say the hosted model falls short
Private model / fine-tuning can start on its own or after an AI production sprint. Contact us for pricing. We fine-tune an open model on your examples only when the evaluations show the hosted model falls short. If it is good enough, you stay on it.
Private model / fine-tuning
An endpoint on Amazon Web Services, Microsoft Azure, Google Cloud or GleSYS, and a simple report that compares the fine-tuned model with your current one on accuracy, cost and speed.
Not a workshop
Why an AI production sprint, not a workshop.
You already know the workflow is worth trying. We put one slice of it into production.
One system
Internal search and answers, a customer or support agent, or an operations agent. Not a roadmap of ten.
Evaluations from the start
The evaluation set and the continuous integration gate are part of the sprint. They are not a later phase.
Senior, in Gothenburg
The people who build it sit with automotive and connected-product teams when the work needs someone in the room.
After the sprint, or instead of it
Where this sits with the rest.
The sprint is one system. Care, a coding setup, a person in the team, or a platform takeover are the other shapes.
Managed AI operations
The retainer that keeps the system healthy after the sprint. New features stay fixed-price change orders.
Read more →Engineering teamsAI coding rollout
When the work is your engineering team's coding agents, not a product agent.
Read more →ConsultingForward-deployed engineers
A senior in your team, for the AI feature and the product engineering around it.
Read more →Full-servicePlatform Partner
When the system lives on an app, portal or connected product that nobody owns yet.
Read more →AIAll AI packages
Sprint, Managed AI operations, AI coding rollout and Private model / fine-tuning, and how we choose between them.
Read more →Related insights
From pilot to production.
How we think about shipping one system and knowing if it still works.
AI pilot to production: a practical playbook for Nordic product teams
Most AI pilots never become product value. Here is a field-tested playbook for taking an AI initiative from demo to stable production impact.
Read insight →AI Product Development · 6 min readAgent readiness before you scale AI
Agentic workflows fail fast when data access, ownership, and logging are unclear. A short readiness sprint surfaces the gaps before production.
Read insight →Questions
Common questions.
Short answers before you book a conversation.
Is this a workshop?
No. It is a build. At the end one system is in your production environment, with evaluations, monitoring and a runbook. We do not deliver a recommendation deck as the result.
What does it cost?
A fixed price, given before we start. It depends on the variant and on where it has to run. Contact us for pricing.
Can you ship three agents in one sprint?
No. One system. The next one is another sprint, or a fixed-price change order once the first is live and the evaluation set exists.
What happens after week six?
You run it, your team runs it from the runbook, or we keep it healthy on Managed AI operations. Scaling is a go or no-go, not an automatic next phase.
Where does the data stay?
In your environment. Private or virtual private cloud on Amazon Web Services, Microsoft Azure, Google Cloud or GleSYS, or on your own servers, is the default, with audit logs. We do not train on your data unless you ask us to, in writing.
Can the model be ours?
Yes. Private model / fine-tuning can start on its own or after an AI production sprint. We fine-tune an open model on your examples only when the evaluations show the hosted model falls short, and we host it on Amazon Web Services, Microsoft Azure, Google Cloud or GleSYS. Otherwise we recommend you stay on the hosted model.
Ready to scope it?
Bring the workflow, not a slide deck
We will tell you straight if it is one sprint, if it should be care of something you already have, or if we are the wrong people.
Ready to talk?
Let's talk about
what you're building
Or reach us directly at hello@aidoni.se
