AI Agents
Agents that take work off the team. Not chatbots that add it.
Most 'AI agents' on the market are wrappers around a single prompt. Real agents take multi-step work — research, drafting, triage, qualification, handoff — and finish them with a result you'd accept from a junior on your team.
What's inside
Model-agnostic. Evaluated before they ship. Monitored after.
- 01
Research and brief agents
Brief in, structured deliverable out. Company research, competitor scans, lead research, pre-meeting briefs. The thing your team keeps not having time to do.
- 02
Inbox and ticket triage
Read, classify, draft a reply, escalate the ones that need a human. Hours back per week per inbox, with the audit trail intact.
- 03
Sales qualification agents
Inbound form to qualified lead in your CRM, with the research attached. No more 'who's this person and what do they want' before every first call.
- 04
Internal-knowledge agents
Your docs, your tickets, your past projects — retrievable in plain English. Built on a real RAG stack, not a quick GPT wrapper.
AI Agent Services
Product-ready agent offers under the AI Agents category.
45,000-95,000 SEK ex VAT
AI Agent Workflows
Scoped agents for support, sales, research, inboxes, and internal knowledge — with logs, evals, approval paths, and handover.
- One narrow use case
- Tool access and permissions
- Evaluation set before launch
Pilot from 19,500 SEK ex VAT
AI Phone Agent
A voice agent that answers inbound calls, qualifies the request, and sends a clean CRM or inbox handoff.
- 10-business-day pilot
- Forwarding-first rollout
- Human handoff rules
35,000-95,000 SEK ex VAT
Phone + Website Agent
One shared intake assistant across phone and website, with channel-specific behavior and one handoff logic.
- Phone and web intake
- Shared context model
- Channel-specific scripts
How an engagement starts
Scope tight. Evaluate hard. Ship in weeks.
- 01
Job-to-be-done
We pick the workflow the agent has to win at. Concrete, measurable, with the success criteria written before any prompt is written.
- 02
Evals before launch
I build the evaluation set first — 30–50 realistic cases. The agent has to pass before it ships to anyone live.
- 03
Production + monitoring
Ship to a small group. Watch the traces. Tune. Expand. The agent gets better in production because someone's actually watching.
Free 30-min review. We pick one workflow worth building.
Common questions
- Which model do you use?
- Model-agnostic by design. I choose the model based on the job, data sensitivity, latency, quality, and cost — Claude, OpenAI, Gemini, local models, or another provider when that is the better fit.
- Do agents replace people?
- Almost never. They take the boring 60% off so people do the 40% that needed a human. The shape of the team changes; the headcount usually doesn't.
- What happens when it breaks?
- Monitoring catches it. The agent fails closed — kicks back to a human — never silently. You see what happened and why, every time.