Question 1

What is an AI agent, and how is it different from a chatbot?

Accepted Answer

An AI agent is software that is given a goal rather than a question: it plans the steps, calls other tools to carry them out, checks the result and repeats until the job is done or it gets stuck. A chatbot returns text and a person does the work; an agent does the work and reports what it did. That difference is the whole reason agents are useful and also the reason they need permissions, limits and an approval step, which a chatbot never does.

Question 2

Is an AI agent the same as a workflow automation?

Accepted Answer

No, and the distinction decides which one you should buy. A workflow follows a path you drew in advance, which makes it predictable and cheap and means it breaks on anything you did not anticipate. An agent decides the path at runtime, which handles messy and varied input but is harder to predict and must be constrained. If your process is consistent and you can draw it, a workflow is the better and safer answer — we will say so rather than sell the more expensive thing.

Question 3

What stops the agent from doing something damaging?

Accepted Answer

Four things, and they are designed in rather than added later: the agent has its own credentials with only the permissions its one job needs; irreversible actions such as sending, paying, deleting or publishing are held behind a human approval step; spend and rate limits cap what any single run can do; and every run is checked against a stated rule, with low-confidence cases escalated to a named person instead of guessed at. Agents given broad permissions and no approval gate are the reason most agent horror stories exist.

Question 4

Which jobs are worth giving to an agent?

Accepted Answer

The ones where input arrives in an inconsistent form and a human currently reads it and decides what to do — triaging inbound email, qualifying inquiries that arrive as free text, reconciling records that do not quite match, researching a prospect before a call, preparing a first-draft response for review. Jobs with clean structured input and fixed rules do not need an agent; jobs where being wrong is expensive and hard to reverse should keep a person in the loop rather than being handed over.

Question 5

How long does an agent build take, and what does it cost?

Accepted Answer

A first production agent typically takes four to six weeks, most of which is spent on scoping the job precisely, setting permissions and building the verification step rather than on the model itself. Pricing is quoted per engagement after a scoping conversation, because the build is sized to your process rather than sold as a package — there is no published price list. Ongoing cost includes model usage, which varies with volume and is billed by the provider rather than by us.

Question 6

What happens when the agent gets it wrong?

Accepted Answer

It should be caught by the verification step, logged, and routed to a person — that path is part of the build, not an afterthought. The realistic expectation to set is that an agent doing genuinely ambiguous work will be wrong sometimes, so the question that matters is not whether it errs but whether an error is visible, reversible and escalated. An agent that fails silently is the failure mode worth designing against, and it is why we shadow the agent against the person doing the job before reducing supervision.

Custom AIAgents

Built to spec.
Ready to deploy.

Scoped to a single job

Connected to your real tools

Approval gates on anything irreversible

Self-checking, with escalation

How it works

Scope the Job

Wire the Tools

Build & Constrain

Shadow & Hand Over

The numbers speak.

Custom AI Agents, answered.

Let's build your
custom ai agents.