Cortex · Guide
Knowledge Agents vs Skills Agents: what's the difference?
By the Uthereal team · July 2026 · 6 min read
When teams start building AI agents on their own knowledge, the first design question is almost always the same: do you need an agent that knows things, or an agent that does things?
It sounds like a subtle distinction. It isn't. It determines what you build, how you build it, and what value you get out. At Uthereal we've made it the core architecture of Cortex: Knowledge Agents and Skills Agents are two different agent types, built for two different jobs.
Knowledge Agents
An expert who has read everything
Answers questions grounded exclusively in the sources you connect — documents, wikis, SharePoint, Notion, Drive, databases, audio, video. Every answer cites the exact passage it came from.
- "What does our contract playbook say about liability caps?"
- "Which of our published studies cover this compound?"
- "How do we configure this product for a dental lab?" — asked by a customer, inside your white-labelled product
Skills Agents
An expert who shows you how the work is done
Doesn't just answer — executes a workflow. Codifies how your best people perform a multi-step task: intake questions, structure, checks, order, and output format.
- Producing a due-diligence brief that follows the firm's exact review structure
- Synthesising new submissions against an editorial taxonomy, flagging gaps
- Drafting the first version of a client deliverable using the house methodology
The one-sentence test
Would a human do this job by looking something up — or by following a procedure?
Looking something up → Knowledge Agent.
Following a procedure → Skills Agent.
And when the procedure requires looking things up along the way (most real expert work does), a Skills Agent uses your grounded knowledge inside its workflow. That's why they live on the same platform.
Why the distinction matters for value
Knowledge Agents defend value you already have. They stop expertise from being trapped in senior heads and 10,000 unread documents. The ROI shows up as speed: faster onboarding, faster answers, fewer interruptions to your experts.
Skills Agents create leverage you didn't have. They take work only your most experienced people could do and make the first 80% of it runnable by anyone — consistently, on every engagement. The ROI shows up as capacity: your seniors spend their hours on the judgement clients actually pay for.
And for consultancies and publishers, there's a third layer: both agent types can be white-labelled and shipped as your own product. Your methodology, packaged as a Skills Agent, becomes something you can license or bundle into delivery — productised IP, not just internal tooling.
Building both, in practice
On Cortex, both agent types come out of the same self-serve workflow — no engineers required.
- Step 1Ingest
Your sources into your private workspace.
- Step 2Ground
A Knowledge Agent on the topics you own.
- Step 3Automate
Expert procedures as Skills Agents.
- Step 4Deploy
To your team, embedded in your product, or via API.
Most teams start with one Knowledge Agent on their most-asked-about corpus, get their experts comfortable, then codify their first workflow as a Skills Agent within a couple of weeks. Time to the first working agent is measured in minutes.
The takeaway
- • Knowledge Agents know things: grounded, citation-backed answers from your sources.
- • Skills Agents do things: multi-step expert workflows, run your way.
- • Real expert work needs both — which is why Cortex treats them as two halves of one platform, not two products.
