Not a chatbot on top of your data
The agent is the architecture, not a feature bolted on at the end. Web, chat, voice, email, Slack and MCP are ways in — one system, one continuous memory behind all of them.
WorkPowers · Copenhagen
AI platforms built on business premises.
We design, build and run AI platforms for organisations that need more than a pilot: agents that do real work, a knowledge layer you can inspect, and a decision trail you can defend. Three of them are in production today.
PathfinderStrategic execution for startups and the programmes backing them
SpeckleLiving specification for software teams and their AI agents
WGLLKnowledge that tells you when to stop trusting it
The premise
Most AI projects don't fail on capability.
They fail on accountability.
A demo persuades in ten minutes. A platform has to hold up for years — through staff turnover, audits, model changes, and the day someone asks why the system said what it said. We build for that day.
Knowledge you can actually see
State lives in an explicit, editable store you own — not in a model's hidden memory. Change vendor or model, and everything you built stays yours, in a form a person can read.
Autonomy with a hand on the wheel
Agents work on their own, propose, revise and retire — up to a confidence ceiling. Certainty and deletion stay with a person. Every step leaves an audit trail you can read back.
What we've built
Three platforms, one spine.
Each solves a different problem for a different buyer. All three stand on the same foundation: structured knowledge, persistent agents, and a traceable line from every claim back to its source.
Pathfinder
Piloting with accelerator cohortsStrategic execution for startups — and for the programmes that back them.
Founders turn ambitious goals into structured, testable plans — goal, outcomes, dependencies, tasks — with a persistent AI agent that pressure-tests the reasoning instead of just filing it. Mentors arrive already briefed. Programme managers get one honest status view across the whole cohort.
What it does
- Breaks a big ambition into what must become true — and what to do about it this week
- An agent that challenges assumptions before the market does
- Check-ins by chat, voice or email: progress interpreted, not merely logged
- Cohort-wide overview — status, momentum, days since last real contact
- Strategy that survives pivots, team changes and the end of the programme
Speckle
In production — running on itselfThe specification that keeps up with the code.
Speckle continuously reverse-engineers a running system into a living, evidence-backed specification — then lets teams and their AI agents question it before they build. Documentation stops being the chore nobody has time for and becomes the thing you consult first.
What it does
- Ask the specification what already exists — from inside your own tools, over MCP
- Agents file what they learn while shipping; people confirm what matters
- Spec branches: change the specification in parallel, merge item by item, on the record
- Onboard onto an unfamiliar codebase — or size up an acquisition target — in days, not months
- A conformance signal showing whether each claim still holds against its cited source
WGLL
Live at whatgoodlookslike.aiKnowledge that tells you when to stop trusting it.
Every entry is one of five kinds — what you hold, what you know, what happened, what you measure, what is still unresolved — and each kind carries its own rule for going stale. Nothing is quietly overwritten; it is superseded. So an overview of your own knowledge is one you can act on.
What it does
- Five classes of claim, each with a maintenance rule built in — no one has to remember to check
- Facts carry their freshness; measures grey out when the cadence lapses
- Standards written as something you can hold work up against, not a slogan
- Areas, topics and links — one record, several readings, no duplicate lists
- Ask your own record a question and get the answer with its provenance attached
How we build
Principles we don't trade away.
These are not aspirations on a wall. They are the reasons our platforms behave the same on day 400 as they did on day 4 — and the first things we'd argue for in your architecture too.
01
Agent-first, not agent-added
The agent is the architecture. Surfaces — web, chat, voice, email, Slack, MCP — are entrances to one system, not separate products with separate memories.
02
The knowledge layer is the product
Structured, inspectable, editable, portable. When a better model arrives next quarter, you swap the model and keep everything you have built.
03
Autonomy with human override
Agents create, propose and retire knowledge up to a confidence ceiling. Only a person confirms with certainty, and only a person deletes for good.
04
Every claim carries its source
Verified against its source, drifted since, or contradicted — stated plainly. A system should never sound more certain than its evidence allows.
05
Earn every interaction
Before the system asks you anything, it has to check whether it already knows. Attention is the scarcest resource in the room, and it belongs to your people.
06
Enterprise reality from day one
Tenant isolation, role-based access, GDPR and EU data residency, encryption, penetration-tested access paths. Designed in at the start, not retrofitted before the security review.
The spine
The same foundation under all three.
Different products, one architecture. It is why a new platform on our stack starts at the interesting part instead of at an empty repository — and why governance is a property of the foundation rather than a feature someone has to remember to add.
Read it bottom-up: the knowledge layer is what survives. Models change, interfaces change, teams change — the structured record of what your organisation knows, decided and measured is the asset, and it stays legible to a human being at every point.
People
Eske Gunge
Founder & CEO
Eske founded WorkPowers and runs it. He is also an angel investor, through Gunge Business and WeShare Invest, and mentors AI startups at Danish accelerator programmes.
That mentoring is where this company's premise came from. He kept meeting teams who had bought AI and still could not say what it had decided, or why — and decided the interesting problem was not the model but everything around it.
Evan Payne
Software Developer
Evan builds the platforms. He works from Barcelona and has been on this work since the first version of Pathfinder — the same hands that wrote the first agent loop still write the code running in production.
Working with WorkPowers
Three ways in.
Use a platform
Pathfinder, Speckle or WGLL, operated by us, with proper onboarding and a named person to call. The fastest route from an idea about AI to something your team is actually using on Monday.
Build on our foundation
We adapt the platform stack to your domain — your methodology, your data, your rules of engagement — so a new AI platform starts from a working spine rather than an empty repository.
Get the operating model straight
A short, hands-on engagement: where agents belong in your organisation, what they must never decide alone, and what has to be true before they are allowed near the real work.
Contact
Tell us what you're trying to build.
The useful first conversation is rarely about technology. It is about the decision you want to make better, and who has to be able to defend it afterwards. Write a few lines about that, and we will tell you honestly whether we are the right people.
- Emaileske@workpowers.ai
- Phone+45 30 63 53 59
- AddressFruebjergvej 3, 2100 Copenhagen, Denmark
- HoursMonday–Friday, 09:00–17:00 CET