Founders who stall on their first AI agent are usually stuck on one question: which job? Everything on your plate looks like a candidate. The choice follows a pattern, though. Score each job on four questions, compare it against the five jobs that usually win, rule out the red flags, and you have your answer in about twenty minutes.
This post is the "which job" decision. Once you have picked, How to hire your first business agent walks through the brief, the accounts, and the first week. For the wider picture, start with Business agents: the complete guide.
Score each job on four questions
List every recurring job you would hand off tomorrow. Score each from 1 to 5 on four questions. Higher is always better; 20 is a perfect first hire.
Hours per week. How much time does this job eat across the whole team? 1 means under an hour a week. 3 means roughly three to five hours. 5 means ten hours or more. Count everyone who touches it.
How repetitive. Does the job follow the same steps most of the time? 5 means nearly every instance looks like the last one. 1 means every instance is a fresh puzzle. Repetitive work is where an agent gets good fast, because each correction applies to every case after it.
How cheap a mistake is. If the agent gets one wrong, what does it cost? Score this one in reverse: 5 means a mistake is cheap and gets caught before anyone outside the company sees it. 1 means a mistake costs real money, a client, or legal exposure. Your first agent will make some week-one mistakes, like any new hire. Pick a job where those are survivable.
How visible the result. Can you check the work in a minute? 5 means the output lands somewhere you can scan at a glance: a calendar, a sent folder, a pipeline table. 1 means you would have to dig to know whether anything happened. Visible work keeps your weekly review short and lets the team see the hire working.
Add the four scores. Anything at 16 or above is a strong first job. Between 12 and 15 is workable. Under 12, keep it for later.
The five usual best first hires
The same five jobs tend to land at the top. Here is when each one wins.
Inbox and email
Wins when the founder's inbox is the bottleneck: an hour or more a day of sorting, routine replies, and "just checking in" threads that hold up everything else. An email agent sorts by what each message needs, drafts or sends the routine replies, and brings you the few that need your judgment. It scores high on hours and visibility. The mistake score depends on how much you let ship on sight, so start with drafts for anything client-facing and widen from there.
Scheduling
Wins when your week is full of external meetings with several people on each side, and every reschedule costs a dozen emails. A scheduling agent handles the back and forth, holds the calendar, and sends the reminders. It scores near the top on repetition and visibility, and a scheduling slip is usually cheap to fix. It is often the safest first hire there is.
Bookkeeping prep
Wins when month-end is a scramble: receipts missing, transactions uncategorized, your accountant waiting on you. A bookkeeping agent categorizes against past patterns, matches receipts, reconciles accounts, and sends the open-questions list before the close instead of after. Note the word prep. The judgment calls in the books stay with you or your accountant. Hours are often modest; this job wins on repetition and a calm close.
Sales follow-up
Wins when leads go quiet because nobody followed up after the first call. This is the job with the clearest line to revenue. A sales agent sends the follow-ups on schedule, keeps the pipeline table current, and flags the deals that need you. Visibility is high, since every touch is logged. Keep pricing conversations and negotiation with a person.
Customer support triage
Wins when you have a product with a steady flow of inbound questions, which usually means a software company or a firm with a client portal. A customer service agent sorts tickets, answers the routine ones, and routes the rest to the right person with the context attached. For a SaaS startup, it is frequently the obvious pick.
Red flags for a bad first job
If a candidate hits any of these, move it down the list.
- You cannot say what done looks like this week. If you cannot describe it, nobody can own it, human or agent.
- It happens once a quarter. Rare jobs give the agent too few chances to learn and you too few chances to correct.
- One mistake is expensive and hard to undo: paying vendors, signing anything, filing with a government agency.
- It needs a license or professional judgment. Tax treatment, legal advice, and audit sign-off stay with the professional.
- It runs on relationships and politics: negotiating a renewal, firing a client, handling a partner dispute.
- Nobody does it well today and there is no example of good. Hand off a process you understand, then improve it.
- You mostly want it gone because you hate it, and it takes twenty minutes a month. Hate is a fine tiebreaker and a poor primary reason.
The best first job is boring, frequent, and easy to check. Exciting comes later.
Worked example: a 6-person consultancy scores its options
Halden & Pike is a hypothetical six-person operations consultancy: two partners, three consultants, and an office manager. They run discovery calls, multi-week client engagements, and a lot of workshops with client stakeholders. The partners scored six candidate jobs in a short meeting.
| Job | Hours | Repetitive | Cheap mistake | Visible | Total |
|---|---|---|---|---|---|
| Workshop and meeting scheduling | 4 | 5 | 4 | 5 | 18 |
| Partner inbox triage | 4 | 4 | 3 | 4 | 15 |
| Sales follow-up after discovery calls | 3 | 4 | 3 | 5 | 15 |
| Bookkeeping prep for month-end | 2 | 5 | 3 | 3 | 13 |
| Client support triage | 1 | 4 | 3 | 4 | 12 |
| Writing first drafts of proposals | 4 | 2 | 1 | 3 | 10 |
Scheduling won clearly. Every workshop involves four to eight people across two companies, and the office manager estimated roughly eight hours a week went to finding times, moving them, and sending reminders. A double-booked slot is annoying and fixable, and the calendar shows at a glance whether the job is done.
Proposal writing eats partner time, but every proposal is different, and a wrong number in front of a prospect is costly. It went on the "later, with a person reviewing every draft" list.
Inbox triage and sales follow-up tied at 15. The partners broke the tie by asking which one they would measure more easily, and picked sales follow-up as the second hire, since days from discovery call to proposal is a single number they already track. Support triage scored low because the firm gets few inbound requests; at a software company it would likely score near the top.
So Halden & Pike hired Iris, a scheduling agent. Iris owns the workshop calendar, sends invites and reminders, and handles reschedules. They bring anything involving a partner's personal time or a client executive to the office manager first.
FAQ
Can my first agent take on two of these jobs at once? They can. Start with one anyway, so you can tell whether the hire is working, then add the second once the first runs on a weekly review.
What if two jobs tie? Pick the one with the higher visibility score, or the one tied to a number you already track. You will learn faster from work you can check quickly.
Should the first job be the one I hate most? Only if it also scores well. A job you hate that takes ten hours a week is a great pick. A job you hate that takes twenty minutes a month is a poor one.
How do I know the first hire is working? Pick one metric before you start: response time, hours back per week, days to close the books, follow-ups sent. Check it in your weekly review. The how-to guide covers that review.
Pick the job, then hire the agent
In Dock, agents are named teammates with jobs, and pricing is per person, per month: Starter at $20, Standard at $100, Scale at $200, each with a monthly allowance of business credits. Your first agent is running in minutes, with no terminal. Start for free.
Part of Business agents: the complete guide.