Here’s a fun exercise: add up every hour you’ve spent this year on hiring, onboarding, or firing someone. Now imagine you’d spent that time on billable work instead. I did this math last month and had to close my laptop and go stare at a wall for a bit. It’s a lot. It’s always a lot.
I’ve spent the last few months writing about becoming a boss by accident — the bad job post, the AI resume triage, the client who somehow became a management problem, the firing I dragged out way too long, the performance reviews I made up because nobody handed me a template. Somewhere around post four I realized I wasn’t just venting anymore. I was building the manual nobody gave me. So here’s the compressed version — the stuff that, if someone had grabbed me by the shoulders and told me before I posted my first “Hiring: Virtual Assistant, Must Be a Self-Starter!!” ad, would’ve saved me probably 40 hours and one very avoidable Slack meltdown.
Lesson 1: Write the outcome, not the job
My first job post was a list of tasks. “Manage inbox. Schedule social posts. Light admin. Some Canva.” Do you know what that gets you? Forty applicants who can technically do those four things and zero who understand why you’re doing any of them. I wrote the whole embarrassing story here, but the short version is: I hired a task-doer when what I needed was someone who understood that “manage inbox” actually meant “make sure I never miss an invoice follow-up again, because I am bad at that and it’s costing me money.”
The fix is stupidly simple and I still have to force myself to do it every time: before you write a single line of the job post, write one sentence that starts with “Three months from now, this is working if ___.” Not a task list. An outcome. “Three months from now, this is working if my inbox has zero unanswered client emails older than 24 hours and I haven’t personally scheduled a social post since March.” That sentence becomes your job post. That sentence becomes your interview questions. That sentence is the only thing you check against when you’re deciding who to hire.
I know that sounds like something a $400 LinkedIn course would tell you to do in a slide with a stock photo of a handshake. But I tested it against my own failure and it holds up. The task list attracts people who follow instructions. The outcome attracts people who solve problems. You want the second kind, always, even for a 10-hour-a-week gig.
Lesson 2: A trial project beats a great interview, every single time
I have now interviewed people who were charming, articulate, said all the right words (“proactive,” “detail-oriented,” “self-starter” — drink every time), and were completely unable to do the actual job. I’ve also hired someone who was awkward on the call, gave short answers, and turned out to be the best subcontractor I’ve worked with, because she asked me three clarifying questions about scope that no one else asked.
Interviews measure how good someone is at interviews. That’s it. That’s the whole skill being tested. So now I pay for a small paid trial task before anyone touches real client work — something that takes them 2-3 hours, that I’d pay for anyway, that mirrors the actual job. For a VA, that might be “reorganize this messy folder of receipts and flag anything that looks off.” For a subcontractor, it’s a scaled-down version of a real deliverable. I pay market rate for it. It is, hands down, the best money I spend in the entire hiring process, because it tells me things no interview can: do they ask questions when something’s ambiguous, or do they guess and hope? Do they hit the deadline, or do they go quiet and reappear four hours late with an apology? You cannot fake a trial project. You can absolutely fake an interview, because I’ve watched people do it to me.
This is also, not coincidentally, where the AI resume sorting comes in handy and where it runs out of road. I used AI tools to get through a stack of 300 resumes for one hire — the story’s over here — and it was genuinely great at the first cut: killing obvious mismatches, flagging keyword overlap, grouping by experience level. What it could not do was tell me who’d actually be good to work with under a deadline. That part’s still just you, a trial project, and your gut.
Lesson 3: Put it in writing, even for the $200 gig
I used to think written agreements were for “real” hires — the ones with contracts and NDAs and all that. Everything else was a DM and a vibe. Then I had a subcontractor deliver something completely different from what I asked for, and when I pushed back, they pulled up our messages and, honestly, they had a point. I’d been vague. I’d said “clean, modern design” and never defined what that meant. Nobody was lying. We just never agreed on what we agreed on.
Now, even for a one-off $150 task, I send a short doc — doesn’t need to be a lawyer’s contract — that covers:
- What “done” looks like, described specifically enough that a stranger could check it off
- The deadline, and what happens if it slips (do you get a heads-up 24 hours early? Do we renegotiate scope?)
- How many revision rounds are included before it becomes a new invoice
- When and how you’re paying them — same day, net 15, on approval, whatever, just say it
Takes me about ten minutes. Has saved me at least three relationships and one very awkward invoice dispute. If you’re already running a team bigger than one, this is basically the same muscle as the feedback structure I built for performance reviews for a team of two — you can’t evaluate someone against expectations that only existed in your head.
Lesson 4: Fire fast and kind, not late and messy
This is the one that costs the most to learn and the one I learned the hardest way. I sat on a firing decision for three months. Three months of knowing, deep down, by week two, that it wasn’t working — and instead of acting on it, I told myself stories. Maybe they need more ramp-up time. Maybe I’m not managing well enough. Maybe next sprint will be different. I wrote out the whole ugly timeline in that post, and the short version is: everyone involved would’ve been better off if I’d made the call at week three instead of week twelve. Including the person I eventually let go — I’d let them keep failing at something that wasn’t a fit for another two and a half months, which is a genuinely unkind thing to do to someone, even though it felt like kindness at the time.
Here’s what I know now that I didn’t know then: the delay is never about them. It’s about you not wanting to have an uncomfortable conversation. That’s it. That’s the whole reason. Dress it up however you want — “giving them a fair shot,” “not wanting to be rash” — but check your gut at week two or three. If you already know, the only thing three more months buys you is a worse version of the same conversation, plus more sunk cost, plus a teammate who’s been quietly stressed about their job security for a season instead of getting clarity in a month. Fast and kind is a real category. It’s not the same as fast and cold. You can give someone two weeks of severance, a genuine explanation, and a same-day decision. What you can’t do is give them false hope for three months and call it compassion.
Lesson 5: AI sorts. It doesn’t judge.
I’ll say this once more because I think it’s the easiest lesson to un-learn under pressure: tools that summarize, rank, and filter are excellent at reducing 300 resumes to 30, or flagging red flags in a contractor’s past work samples, or drafting the first pass of a job post. They are not good at deciding who you should trust with your clients’ money, your reputation, or your Tuesday afternoons. I leaned on AI hard for the resume grind and don’t regret it — it turned a two-week slog into an afternoon. But every hire I’ve regretted is one where I let the efficient part of the process (sorting, scoring, shortlisting) quietly stand in for the part that actually requires me: talking to a human, running the trial, trusting or not trusting my gut. Use the tools to get to the decision faster. Don’t let them make the decision for you.
Same goes for juggling the humans once they’re hired — when I was drowning in conflicting priorities from different clients and, later, different team members, no tool sorted that out for me either. I had to build an actual system for it, which I wrote about in the multiple-bosses post. Software can hand you the inputs. It can’t tell you whose fire is actually the biggest one.
The part nobody tells you
When I quit my job to freelance, nobody handed me a manual for being my own boss. I figured it out by underpricing my first ten projects, missing a quarterly tax payment, and slowly building a system out of the wreckage. Turns out becoming boss of other people works exactly the same way. There’s no manual. There’s just you, a series of avoidable mistakes, and — if you’re paying attention — a slightly better system after each one.
I don’t think that’s a design flaw, honestly. I think anyone who tells you they had management figured out on day one either had a business school beat it into them or is lying to sound impressive at a networking event. You’re allowed to be bad at this at first. You’re allowed to write a bad job post, sit on a firing too long, and vaguely gesture at “clean, modern design” and hope for the best. The only real failure is not updating the system after it breaks. Write the outcome down. Run the trial. Put it in writing. Act on what you already know. That’s the whole manual. I wish someone had handed it to me sooner, so now you’ve got it a little earlier than I did.