Short answer
Ask four questions about the work: how often it happens, how much judgement it needs, how bad a mistake is, and whether anyone has written it down. Frequent and rule-shaped goes to an automation. Frequent with variation goes to an agent. Judgement-heavy or relationship-bound stays with a person.
Key takeaways
- The wrong choice is usually hiring for work that was never designed, or automating work nobody has documented.
- Undocumented work cannot be given to software at all. Writing it down is the first deliverable either way.
- Cost is not the deciding factor. Reversibility is: how visible and how expensive is a mistake.
- Most real answers are combinations, with the agent drafting and a person approving.
The instinct when a team is drowning is to hire. Sometimes that is right. Often what is actually happening is that nobody has ever written down how the work is done, so each new person learns a slightly different version of it, and the problem scales with the headcount.
The four questions
How often does this happen: daily, weekly, or a few times a year? How much judgement does a good outcome need? If it goes wrong, does a customer see it before you do? And has anyone written down how it is done today? The fourth question decides whether you can act on the first three at all.
| The work | Best fit | Why |
|---|---|---|
| Daily, identical every time, low stakes | Automation | Cheapest, fastest, and it fails visibly |
| Weekly or daily, same shape but details vary | Agent, with approval | Handles variation that breaks rigid rules |
| Frequent, but a mistake reaches a client directly | Agent drafts, person sends | Keeps the speed, keeps the judgement |
| Judgement, relationships, negotiation | Person | The part of the job that is actually the job |
| Rare, a few times a year | Person, with a checklist | Never enough repetitions to make software reliable |
| Nobody can describe how it is done | Document it first | You cannot delegate what has never been defined |
Why documentation is the real gate
An agent can only follow a process that has been described, and most processes live in one person’s head. When that person writes it out, two things happen: half the steps turn out to exist for reasons that expired years ago, and the remaining half get simpler. Simplify first, then automate what is left.
This is also the honest answer to why two quotes for the same build differ by a factor of five. A business with written procedures is buying an installation. A business without them is buying a discovery project first, and then an installation.
What hiring still solves better
Judgement under ambiguity, relationships that carry revenue, work that requires physical presence, and anything where accountability must sit with a person. Also growth: a good hire changes what the business can attempt, which is a different question from whether this week’s tasks get done.
The combination most businesses end up with
An agent that prepares and a person who decides. It drafts the follow-up, assembles the reconciliation, checks the output against the standard, and a named human approves what goes out. You keep the speed of software on the preparation and the judgement of a person at the point where a mistake would cost you.
Then, once the quality is proven over weeks rather than demos, you move the approval point on the lowest-risk slice and watch it for another few weeks. That is the whole method. It is unglamorous and it is why the installs survive.
Questions people ask
Is an agent cheaper than hiring someone?
Usually in direct cost and never in zero cost. There is a build, then model usage, subscriptions, oversight and periodic rework. The better comparison is what each option can absorb: a person handles ambiguity, an agent handles volume without complaining.
Can I automate work my team has not documented?
No, and attempting it produces a faster version of the wrong process. Have the person who does the task write it out, including the decisions they make without noticing. Expect to find steps that exist only because someone left years ago.
What if the work is somewhere in between?
Split it. Give the repeatable preparation to an agent, keep the decision with a person. Most durable setups look like this rather than a clean handover, because the preparation is where the hours are and the decision is where the risk is.
Should I automate before hiring?
Document before either. Once the process is written down, the right choice is usually obvious, and whichever you choose works better. Hiring into an undefined process is how you end up with three people doing the same job three different ways.
What is the smallest sensible first step?
Pick one workflow that happens at least weekly, where two people would do it the same way and a mistake is caught before a customer sees it. Write it down, build one agent for it, and measure hours and error rate for four weeks.
Sources
- delegAIte published pricing and Paths In, delegaite.co (read 20 September 2026): That the three routes into this work are published with prices rather than quoted privately.
Published . Last reviewed .
Want this in your business?
Book a call and we will scope exactly which part of your operation an agent should take first.
Book a call with the team