The four options, side by side
Each row compares one thing buyers check.
Hire a person, give the team ChatGPT, build it yourself, or use Surehand on Gatehouse.
ALL FOUR ARE SOMETIMES THE RIGHT ANSWER.
A hire, or overtime on the team you have.
A chat window, and someone pasting into it.
Your own engineers, AI models and on-call rota.
We build and run governed agents for you on Gatehouse, our control plane.
Each row compares one thing buyers check.
| Hire a person | Give the team ChatGPT | Build it yourself | Surehand (us) | |
|---|---|---|---|---|
| Time to first result | A hiring round, plus notice | Same day | Months | Set in the teardown |
| How cost grows | Salary, fixed and rising | Flat, per seat | Large up front, then upkeep | Agreed in the teardown |
| Copes with more work | No. You hire again | Only as fast as people type | Yes, while maintained | Yes |
| Runs unattended | No | No. A person has to paste | Yes | Yes |
| Handles odd cases | Yes. That is the job | Yes, if asked well | As far as you coded for | Yes, or it holds |
| A record you can hand an auditor | Notes, if any were kept | No record you can hand over | Logs, which seldom count as evidence | A sealed record, with evidence and approver |
| A hard limit on spending | Policy and judgement | None | Only if you built one | Checked before each action runs |
| When it is not sure | They ask someone | It answers anyway | Only if you built a hold | It holds for one named approver |
| Who is accountable | The person | Nobody, in practice | The team that built it | Named in the record |
| If it ends | What they knew leaves with them | Nothing to keep | You keep it all, and the upkeep | Export in open formats, reads without Gatehouse |
| Where it breaks first | Volume, and holidays | Consistent answers, and proof | Upkeep, once the build is done | A process nobody wrote down |
The last row shows where each option breaks first.
We cannot give you a number without knowing your volume. We can show how each cost grows. A hire adds cost one whole person at a time, and that cost stays the same in a quiet month.
Which option costs less depends on how many cases need a person. The teardown measures that, along with your volume. For more, read How much does an AI agent cost?
For a lot of work, one of these three is the better choice. Here is when to pick each one.
The value is that a human is on the other end, as with key accounts, negotiations and delicate talks. An agent would lose that. Hire too if the work fits inside one person's day and is not growing.
You want help with drafts, summaries and ideas, and a person reads each result before it is sent. That works well, costs little and is ready today. It stops being enough when work must run with nobody at the keyboard. It also falls short when you must later prove what happened and why.
Build it if the decisions you want to automate are your product. You should already run live software with engineers on call, and at least two people should own it. Then build it with the controls set out in What is AI governance? None of it is secret, and we want you to do it well.
Surehand fits when all four of these are true.
If all four are true, an agent can do this work under rules. If any one is false, one of the options above is likely the better choice. Surehand builds and runs the agents, and they run on Gatehouse, the product we make.
Work we turn down, so you can rule it out before the teardown.
The policy check adds a little time and leaves a record. If the answer must arrive before a web page loads, use something else.
Write the process down before you automate it. Do not let anyone sell you automation for a process that lives only in someone's head.
A hold stops the cases your rules flag, so a mistake nobody wrote a rule for can still get through. Where one mistake cannot be undone, keep a person doing the work.
If the process takes little time today, the teardown may cost you more attention than it saves. Come back when it grows, or when you need to prove what happened.
Sometimes. It takes over the decisions in a role that repeat, such as matching invoices or drafting the same reply from the same evidence. It does not replace a person whose value is the relationship, the negotiation or a call that needs fresh judgement.
A chat window needs a person at the keyboard and leaves no record you could hand an auditor. It also has no limit on what it will say. A governed agent runs on its own inside a signed manifest on Gatehouse. It cannot go past its spend limit. Where judgement is needed, it holds for one named approver. Each run is sealed into a record with the evidence it used. Both tools are useful, for different work.
You take the audit trail, policy history and labelled examples in open formats. They read without Gatehouse installed, so you do not need us to open your own records.
Tell us what comes in, who handles it and where it goes wrong. A person replies, usually inside two business days.
support@surehand.io