AI Automation for Existing Tools | Kingfisher Software

Q: Where does the agent run?

On your infrastructure or on a managed instance we maintain. Either way, the agent has scoped credentials to your repos and project tracker, and nothing else.

How it works in practice

Three loops we set up most often. All run on your repos and your project tracker. Nothing leaves your stack except the model calls.

~30 min vs ~2 days

Bug-fix loop

File a bug as an issue with a screenshot or stack trace. The agent reads it, traces the cause through the codebase, and opens a PR. You review and merge. Same as any other PR.

3-5× faster

Routine feature delivery

Add an auto-implement label to a well-scoped ticket. The agent scaffolds, writes tests, and opens a PR. You spec, review, ship. Best for CRUD, integrations, and anything pattern-matched against existing code.

Hands-off

Self-maintaining tools

Internal dashboards and admin panels that update themselves as your data shape changes. No more "can someone please add this column to the report". The agent picks up the schema change and PRs the UI.

Engagement tiers

Same fixed-scope, fixed-price model as everything else we do. Start with an audit, scale up if it works.

Discovery audit

€1,500-2,500 1 week

2-3 hour workflow review, three concrete automation candidates ranked by ROI, and a written recommendation. No commitment to build anything afterward.

Pilot integration

€8,000-12,000 2 weeks

One workflow, one repo, working agent loop deployed to your infrastructure. Includes 2 weeks of post-launch tuning support and a knowledge-transfer doc your team can run with.

Full deployment

€25,000-45,000 3-6 weeks

Multi-workflow setup, custom prompts and tooling, audit logging, role-based safeguards, 30 days of post-launch support, and a runbook for the on-call surprises.

Retainer (optional)

€2,500-6,000 / mo ongoing

Tuning, new use cases as they come up, and on-call when the agent does something unexpected. Cancel anytime, no minimum commitment beyond the current month.

What this is not

A few clarifications, since "AI" gets used to mean a lot of different things.

Not a chatbot you talk to. The agent reads tickets and writes code; you review PRs.
Not Cursor or Copilot for your team. Those are editor tools. This runs autonomously on your tracker.
Not generic AI strategy consulting. We build the working system, not slide decks.
Not a black box. Runs on your infrastructure, every change is a normal PR, code is yours.
Not a replacement for engineers. Closes the routine loop so engineers spend time on real work.
Not bespoke model training. We use frontier models (Claude, GPT) with engineered context, not finetunes.

Frequently asked

Where does the agent run?

On your infrastructure or ours, your call. We can deploy it to your existing Kubernetes / VPS / cloud, or run it on a managed instance we maintain for you. Either way, it has scoped credentials to your repos and project tracker, and nothing else.

What about API costs at scale?

Frontier model API calls are billed pass-through at our cost, no markup. A typical busy customer spends $200-$1,200/month on tokens. We set a soft cap in the contract so you can't get surprised by a runaway loop. Caching and prompt design keep this number boringly low.

Do you keep our code or data?

No. The agent has read access to your repos to do its job, but we don't store or train on anything. Logs of agent actions stay on your infrastructure. We sign an NDA before the discovery call if you want one.

Can the agent break production?

The agent only opens pull requests. It doesn't push to main, doesn't deploy, doesn't run migrations. Your existing CI/CD and review process gates everything. Audit logs record every action.

What if our codebase is messy?

Honestly, that's most codebases. The agent works best when it has clear conventions to imitate, but it's not blocked by mess. It reads the code as it is. Part of the discovery audit is figuring out which parts of your codebase are in good enough shape for the agent to work in.

Can we self-host after you build it?

Yes. The full deployment tier includes a runbook so your team can take over operations. The retainer is there if you want us to keep tuning it; it's not required.

AI automation for your existing tools.