Everyone has the tools. Few have changed how they work.
Usage is uneven. A handful of people run agents in parallel. Most of the team stops at autocomplete. The gap grows every month.
Agents fail on context, not intelligence.The model doesn't know your conventions, your last incident, or which doc is current. So it guesses.
Spend grows and nobody can explain it. Seats, API keys and side subscriptions multiply. Nobody owns the bill, the data policy, or the call on which tool stays.
AI-native means your workflows, knowledge and guardrails are built for agents from the start. That's what we build with you.
Four things we change.
Most engagements touch all four. You can start with any one.
Level up your team
- Before
- A few power users. Everyone else stuck at autocomplete.
- After
- Every engineer runs agents on real work, with shared rules for what good looks like.
- Hands-on workshop series, run on your own codebase
- Agentic engineering guidelines written with your team
- An AI-first hackday to get everyone shipping
- Adoption survey and baseline, so progress is measurable
Put AI in the workflow
- Before
- AI lives in a chat tab. The work happens somewhere else.
- After
- Agents triage issues, review pull requests, keep tickets current and answer team questions. A person approves each result.
- Bug triage agent that drafts the root cause and the fix
- AI code review on your infrastructure and your models
- Agents that keep tickets and documentation current
- Slack assistants that answer from your own sources
Give AI your company's context
- Before
- Knowledge is spread across Slack, Jira, Drive and GitHub. Agents see none of it.
- After
- One context layer, kept current automatically, that every agent and AI tool can read with the right permissions.
- A knowledge base synced from the tools you already use
- An MCP server, so any AI tool can query it
- Access control per person and per query
- Freshness rules, so a stale doc never wins
Run it safely, at a known cost
- Before
- Five AI vendors, no view of spend, no data policy.
- After
- One model gateway, clear budgets, tracing and evals. You decide which model runs where, and what it costs.
- Model gateway with routing and zero data retention
- Cost dashboard, budget caps and Slack alerts
- Sandboxed runtime for background agents
- Tracing and evals on open tooling, without lock-in
- Tool consolidation and build-or-buy decisions
How an engagement runs.
A typical timeline. We scope it to your team's size and starting point.
- Weeks 1-2
Assess
We survey your team, review your tools and spend, and sit in on how work gets done. You get a baseline and a ranked plan.
- Weeks 2-6
Enable
Workshops on your own codebase. Guidelines your team writes with us. Tool decisions made and rolled out.
- Weeks 4-12
Build
We ship the first agents into real workflows, with the context layer and guardrails they need to work.
- After
Hand over
Your team owns it. Runbooks, dashboards and evals stay behind. We stay on call for review when you want a second pair of eyes.
Done before, inside a real company.
Built while leading AI platform work at a global developer-platform company. Same playbook, now for your team.
An SDK prototype in two weeks
A new Flutter SDK, from empty repo to working MVP in about two weeks, built by running several agent sessions in parallel.
A triage agent for customer issues
Reads each incoming issue, pulls context from code, tickets and Slack, and drafts a root cause. Engineers approve. Next step: the agent opens the fix.
AI cost governance from scratch
A spend dashboard, per-person budgets, a tiered ladder of Slack alerts and hard caps. Published spending guidelines so everyone knows the rules.
A company-wide context layer
Markdown kept current from Jira, Slack, Google Drive, GitHub and AWS, served to any AI tool over MCP.
Guidelines, workshops and a hackday
Co-wrote the company's agentic engineering guidelines, ran a Claude Code workshop series and an AI-first hackday.
Tool consolidation and ROI
Ran the evaluation between two enterprise AI coding tools, drove the decision, and reported productivity impact to leadership.
How we build.
Trust the scaffolding, not the model.
A weaker model in a tight harness beats a stronger model in a loose one. We build the harness: scoped tools, clean context, explicit checks.
Your context is the asset.
Models are rented and get replaced every quarter. Your knowledge, rules and evals are what compound.
Evals from day one.
An agent without evals regresses quietly. Evals are the test suite for agents, so they go in before the first launch.
A person signs off.
Agents draft. People decide, approve and own the result. Every workflow we build makes that step explicit.
No lock-in.
Open, self-hostable tools where control matters. Swap models or vendors without rebuilding what sits on top.
You work with the person doing the work.
Nurbot is founded by Tobrun Van Nuland.He spent more than ten years building SDKs and platform infrastructure at a global mapping company. In 2023 he moved into production AI full time, founded the company's AI platform team, and built its agent runtime, context layer and cost governance.
He also runs a self-hosted inference lab with LLM servers and computer use sandboxes. Knowing the stack from the silicon up is how we give straight answers about cost, privacy and what runs where.
Start with a 30-minute call.
Tell us where your team is today. You leave with three concrete next steps, whether or not we work together.