AI agents vs AI assistants for investment teams.
The industry uses the words interchangeably. They should not. An assistant helps a human do work. An agent does work on behalf of the firm — with tools, scope and an audit trail.
A scoped, autonomous operator inside an AI OS that can call tools and take actions on behalf of the firm within a defined policy, with a full audit trail of every decision, tool call and output.
An assistant answers questions. A human reads the answer and decides what to do. An agent takes actions. It calls tools, queries systems, drafts documents, escalates when it hits its scope. The difference sounds small. In production it is the difference between a chatbot that saves five minutes and an operator that runs a workflow end-to-end.
What makes an agent an agent
- Tools: a defined set of functions it can call — search Firm Memory, query a system, write a draft, book a meeting, open a ticket.
- Scope: an explicit policy for what it may and may not do, and when it must escalate to a human.
- State: it holds context across turns, so a multi-step workflow does not restart every message.
- Audit: every decision, tool call, input and output is logged and reviewable.
- Evals: it is measured against a ground-truth set the firm owns — pass rate, hallucination rate, escalation rate.
Why financial firms need both
Assistants are right for tasks a human still owns — a PM reviewing a company, an adviser prepping for a client call. Agents are right for tasks the firm wants to industrialise — every inbound complaint triaged the same way; every filing tagged, summarised and routed the same way; every new client's KYC pack assembled the same way. Firms confuse the two, buy a wall of assistants, and wonder why productivity did not move.
Assistants save time. Agents change headcount. The OS should ship both, deliberately.
Where to start
Every KJ Capital engagement begins with a two-week £15k AI Diagnostic. We map your existing data estate, model your firm's OS across the six layers above, cost the build, name the first three skills to ship, and hand you a plan your board can act on. It is deliberately priced to be a rounding error against the value it unlocks.
FAQ
Aren't agents too risky for regulated firms?
Only if built without scope, evals or audit. A well-scoped agent — with narrow tools, forced escalations and full logs — is more auditable than the humans it replaces because every action is recorded.
What tasks are wrong for an agent?
Anything the firm cannot write down as a policy. If a senior human cannot articulate 'here is exactly how we do this and when we escalate', an agent should not do it either.
How many agents does a firm typically run?
Five to fifteen in the first year. Each one owns one workflow. Agents proliferate when a firm confuses them with skills — those should scale to dozens, agents should stay lean.
Start with the £15k AI Diagnostic
Two weeks, one costed OS blueprint, three named skills to ship first.