Home / AI Product Engineering / AI agent development
AI Product Engineering · Agents

AI agent development, with a person in control.

We build agents that read your systems, prepare the work and ask before they act. We run 50 of them ourselves, from code review to invoice reminders, under the same rules we build for clients.

Talk about an agentSee how we build

A 30-minute call with an engineer who builds AI systems.

We have built
50 agents for sales, delivery, support and finance
Connected through
MCP servers with only the tools each agent needs
Control
Approval steps and per-user permissions
Every run
Logged with its tokens and cost
What we build

Agents for real work

Back-office agents

Draft quotes, chase late invoices, reconcile payments and update records, ready for review.

Support agents

Triage tickets, draft replies from your knowledge base and route the hard cases to people.

Delivery agents

Plan sprints, review code, write release notes and report blockers inside your project tool.

Multi-agent workflows

Several agents with one job each, coordinated by a dispatcher with budgets and approval gates.

MCP servers

Safe connections between AI models and your systems, loading only the tools a task needs.

Guardrails

Permissions, approval queues, spending caps and an audit log of every action and refusal.

Our own agents

Agents we run every day

Crux by RedmineFlux · 25 agents

Project delivery

Code review, root-cause analysis, sprint planning, helpdesk triage and docs, coordinated with a run ledger of tokens and cost.

Sales · 13 agents

Proposals and replies

Lead qualifying, proposal drafts and email replies. The sales team reviews every one before it goes out.

Odient · 3 agents

Inside Odoo

Ask Odient, Diana in Discuss and live insights, drafting work for approval across 15 Odoo modules.

How we build

How an AI agent development project runs

01
With your team

Pick one task

We choose a task that is frequent, rule based and easy to check.

You getA clear first target
02
Before any code

Write the rules

What the agent may read, what it may prepare and who approves it.

You getAn approved agent spec
03
Only the tools needed

Connect your systems

We build or reuse MCP tools for the systems the agent needs.

You getLeast-privilege access
04
Before go-live

Test on real cases

First, the agent drafts work on past cases. Then your team checks the results.

You getMeasured accuracy on your data
05
Tuned as it runs

Run with approval

Finally, it goes live with approval steps, budgets and a full log.

You getAn agent in daily use
Case studies

Related client work

Senior living · USA

OpenAI chatbot that helps seniors avoid fraud

Older people needed one trustworthy place to learn about Medicare, Social Security and financial fraud, in plain language.

Chatbot in five languages, with voice
2 months · 360 hoursRead the case study ↗

See all AI case studies

Other AI services

Related services

AI agent developmentAgents with approval and auditGenerative AI & LLMChat and search on your dataAI for existing softwareAdd AI to what you runWeb developmentWebsites, stores and web appsMobile appsiOS and Android with FlutterCustom softwareBuilt around your process
FAQ

Common questions

What is an AI agent?

An AI agent is software that uses a language model to work through a task in steps: it reads data, uses tools and prepares a result. Ours stop before any action that changes your data and wait for a person to approve it.

How do you stop an agent from making mistakes?

First, each agent has one job, access to only the tools it needs, a spending cap and an approval step. Also, we log every run, so you can see what it did and why.

Which systems can agents work with?

Anything with an API. We have built MCP servers for Odoo, Redmine and WordPress, and we connect to other systems the same way.

How long does AI agent development take?

A focused first agent for one task is usually a matter of weeks. Then we agree the scope and timeline in the spec before we build.