Fix-It Sprint
1–2 weeks
- Rapid assessment, urgent fixes
- Intensive, one problem
week 1week 5week 8
How to hire us
Hands-on · for real systems under pressure
Anti-consulting.
Your AI isn’t failing because of strategy decks. It’s failing because real systems are messy.
A strategy slide gets struck out. In its place: the system’s logs, the causes we find in them, the fixes, and the numbers before and after.
01 · When teams call us
For systems that
already hurt.
Anti-consulting means we open the logs before we open a slide deck.
When teams call us
The AI system is live and something is wrong. Answers are inconsistent, the API bill doubled, users stopped trusting it, or security is blocking the next release. The team is too close to it, too busy, or has already tried the obvious fixes.
What we do first
We read what is really there: code, prompts, retrieval setup, data, logs, costs. Most problems come down to a few causes: retrieval that returns the wrong context, prompts that grew by accretion, no evals, the wrong model for the job, or missing guardrails. We find the highest-leverage fixes and implement them, with numbers before and after.
Then what
Deeper changes where they are needed, and knowledge transfer so the team can continue without us: training on what we changed and why, and runbooks for the next incident. As a monthly retainer, embedded part-time in your team, or as a one- to two-week fix-it sprint.
- QualityThe answers are bad.
Inconsistent or wrong outputs, and nobody can say how often.
- CostThe bill is too high.
API costs outpace the value, and usage keeps growing.
- TrustNobody believes it any more.
Users route around it. Stakeholders ask whether to switch it off.
- SecurityLegal or InfoSec says no.
The next release is blocked, and the questions have no written answers.
- TeamThe engineers are split.
Two camps, two approaches, and nothing moves while they argue.
- Black boxNobody can explain it.
It does something different every week, and nobody knows why.
No slides. No buzzwords. Real fixes, in real systems.
- Fastest
- Fix-It Sprint, 1–2 weeks
- Ongoing
- Retainer or embedded
- You get
- Fixes with before and after numbers
02 · Where it breaks
Six places AI
systems break.
Most problems come down to a handful of causes. Pick one to see what we look for and what we change.
Retrieval & knowledge
RAG, embeddings, search
- What we see
- Answers cite the wrong document, or none. The right one exists but never gets found.
- What we change
- We test retrieval on its own, against the questions your users actually ask: chunking, embeddings, hybrid search, reranking.
- What you get
- A retrieval score you can track
03 · Process
Assess. Fix.
Improve. Hand over.
Quick wins first, so the pressure drops. Then the deeper work. Then the knowledge moves to your team.
- 01System assessment
We read code, prompts, data, retrieval, logs and costs, and rank what we find by impact.
- Output
- Assessment report and priority issues
- We need
- Access
- 02Quick wins
The highest-leverage fixes, implemented and measured, so everyone sees the difference.
- Output
- Implemented fixes with metrics
- We need
- Fast decisions
- 03Systematic improvement
The deeper changes: architecture, evals, monitoring, whatever the causes call for.
- Output
- System improvements and documentation
- We need
- Engineering time
- 04Knowledge transfer
What we changed and why, so your team can carry on and handle the next incident.
- Output
- Training sessions and runbooks
- We need
- The team
04 · Engagement models
A sprint, a seat,
or a standing call.
Same people, same way of working. What changes is how often we are in your system.
Embedded Consultant
part-time
- In your team, in your repo
- Hands-on sessions, pair programming
week 1week 5week 8
Office Hours Retainer
monthly
- Ongoing advisory
- Architecture reviews
week 1week 5week 8
The strips show an illustrative rhythm. The real one is agreed per engagement.
05 · Self-check
What kind of help
do you need?
Four questions. The result is the engagement that fits, which is often the smallest one.
result0 / 4 answered
- Fix-It Sprint.
One to two weeks on one problem, with before and after numbers.
- Office Hours Retainer.
Monthly advisory for decisions and reviews while your team builds.
- Embedded Consultant.
Part-time in your team and your repo, for as long as the system needs it.
01What hurts most?
02Does your team have capacity to implement?
03How much will the system keep changing?
04How often do you need to talk decisions through?
06 · Fit
Call us if.
Don’t if.
Consulting starts where a system is already live. Before that, other work fits better.
A good fit
- Your AI system is in production and underperforming.
- Costs are spiralling.
- Users or stakeholders stopped trusting it.
- Security or legal concerns are blocking progress.
- Your team is stuck and needs an outside view.
Look elsewhere
- You need a strategy and a roadmap. That is AI Strategy.
- You need a full team to build something new. That is Implementation.
- You still need to decide what to do first. That is a Workshop.
07 · Questions
Before
you call.
What does anti-consulting mean?
We read the system before we talk about strategy: code, prompts, retrieval, data, logs and costs. The output is fixes with numbers before and after, not a deck.
Do you advise, or do you change the code?
Both. We implement the fixes in your repository, measure the effect, and explain what we changed. In the embedded model we pair with your engineers on the work.
What access do you need?
To start: read access to code, prompts, the retrieval setup, logs and cost data. Where we implement fixes, write access on your terms, through your normal review process.
Which areas do you cover?
Retrieval and knowledge systems, tool calling and agent design, evaluation harnesses, prompting and output control, deployment and monitoring, and training for the team. The system map above shows where each one sits.
We need a strategy, not fixes.
Then AI Strategy fits better. Consulting starts where a system is already live and hurting.
08 · How to hire us
Three ways in.
Pick by where you are: deciding, stuck with a live system, or ready to build.
01 / 03Workshop
Half or full day. Ends with decisions, owners and next steps.
When you need to decide
02 / 03you are hereConsulting
03 / 03ImplementationRetainer, fix-it sprints, embedded help. For systems that are live and hurting.
When a live system hurts
Two- to eight-week builds. From validated idea to real users.
When you are ready to build
What is breaking in production right now?
Tell us
what’s stuck.
One call. Bring the symptoms, we bring the questions. If it isn’t a fit, you’ll hear that too.