Can't AI Just Do It in a Day?

Have you ever felt replaced by a tool you were the one using?

A few weeks ago I was on a planning call. We'd just walked through a feature: auth changes, a couple of new endpoints, some migration work. Solid two-sprint job. Then one stakeholder leaned into the camera and asked, very sincerely:

"Can't AI just do this in a day?"

Silence. About three seconds. In meeting time, that's long enough for NEPA to take light and restore it.

I looked at the developer who'd be building it. They'd gone quiet, the kind of quiet where you can almost hear someone updating their CV. Or checking Canada visa requirements. And I recognised it, because I've had the same thought: if the answer is yes… wetin I come dey do here?

I sit in an odd spot. I'm a Technical Project Manager, so I build the plans, make the estimates, and answer "when will it ship?" about 40 times a week. But I also write code, and I use AI tools daily. I've been the one setting the expectations and the one sweating to meet them.

So here's how it looks from both sides of the sprint board, from here in Nigeria.

It's not one anxiety. It's three in a trench coat.

We say "AI anxiety" like it's one thing. In practice I see at least three, stacked on each other and pretending to be one adult.

1. "Will there be a job for me?"

The big one. It lives in every layoff headline and every "AI will replace X by next Tuesday" thread. For many of us here it hits differently, because a lot of Nigerian devs work remotely for companies abroad. When a startup in San Francisco or Berlin decides to "do more with AI," the cut often reaches the remote contractor in Yaba or Abuja first. It's the anxiety you control least, so it shows up as a background hum, like your neighbour's generator at night.

2. "Am I still getting better, or just prompting?"

Quieter, and probably more common among devs who use the tools heavily. You ship more, but you're not sure you understand more. You accept a generated function, it passes, you move on, and a small voice asks: if this thing breaks at 2am, you fit debug am?

(The small voice is annoying. The small voice is also usually right.)

3. "The bar for fast went up. My hours didn't."

This is the one I want to spend time on, because it's the one we create ourselves.

The PM confession

Here's something people in my role rarely say out loud:

Estimates are quietly shrinking because "AI helps," and nobody is checking whether that's true.

It's rarely a formal decision. A 5-point task becomes 3 "since Copilot can scaffold most of it." A deadline moves in because "the team has AI tools now." Someone watches a demo of an agent building a full app in ten minutes and recalibrates their whole worldview.

Here's how it actually plays out, in a highly scientific table:

Task	Original estimate	"AI-adjusted" estimate	What actually happened
Write the feature	3 days	1 day	1 day ✅
Explain the codebase to the AI	not a thing	not a thing	0.5 days
Review code you didn't write	included	"it's fine"	1 day
Fix the bug that looked correct	—	—	1.5 days + one existential crisis
Total	3 days	1 day	4 days

The demo never shows the review, the verification, the context-setting, or the subtle bug in code that was confidently, beautifully wrong. It also doesn't show the hour you lost when light went off mid-session and your inverter decided today was not its day.

AI does speed up some work. I've seen it. But it also creates work, and that new work usually isn't on the board. When the plan banks the speed-up and ignores the overhead, the gap doesn't vanish. People absorb it: their evenings, their weekends, and the creeping feeling that everyone else is keeping up except them.

That's where it stops being a planning problem and becomes a mental health one.

What's actually true (as far as I can tell)

I'm not going to tell you AI will never replace developers. I don't know that, and neither does anyone posting confident threads about it with rocket emojis.

But I'm also not going to tell you to "adapt or die," because that's not insight. It's fear wearing a hoodie and a tech-conference lanyard.

Here's what I'm fairly confident about:

• The tools are genuinely useful. Pretending otherwise helps nobody.
• The bottleneck is moving, not disappearing. Writing code is getting cheaper. Knowing what to build, why, and whether the result is actually correct is not.
• Judgment is getting more valuable, not less. System design, debugging, domain knowledge, explaining trade-offs. These were always the hard parts. Now they're visibly the hard parts. And local context matters: an AI tool won't automatically understand why your payment flow needs three fallbacks or why your app has to work on patchy 3G.
• Uncertainty is the real stressor. Much of the anxiety isn't about what AI does. It's about not knowing what leadership plans to do about it.

That last point matters, because uncertainty is something teams can actually reduce.

What helps: for individuals

Keep some deliberate "no-AI" practice. Not as a purity test, but as maintenance. Debug something by hand. Build a small feature from scratch. Actually read the code before you hit accept. Pilots still fly manually sometimes even though autopilot exists, and nobody calls them old school.

Invest in skills that compound. Debugging, architecture, reading unfamiliar codebases, writing clearly. These get more valuable the more AI-generated code exists. Somebody has to understand the spaghetti.

Be smart about the tools you pay for. Many of these subscriptions are priced in dollars, and we're earning in naira. You don't need every tool your timeline is hyping. Pick the one or two that actually save you time and learn them properly.

Say the fear out loud. To a teammate, your manager, your guys in the industry. I've been surprised how often "honestly, I'm worried about this" in a 1:1 gets "…me too." Anxiety shrinks when it's shared. It grows when it's hidden.

Compare yourself to your past self, not to a demo. Demos are curated, rehearsed, and probably recorded on the fourth try, with stable power and fibre internet. Your codebase has a file called utils_final_v2_REAL.js. Different leagues.

What helps: for teams and the people who plan them

This bit is mostly for my fellow PMs, tech leads, and managers.

Don't cut estimates on assumed AI gains. Measure first. If the team is really faster after a few sprints, adjust then. Don't spend the speed-up before it exists.

Put review and verification on the board. If reviewing AI output is real work (it is), make it visible work. Invisible work is where burnout hides.

Push back on "can't AI just do it?" Kindly, with specifics. Bring the table above if you have to. Part of our job is protecting the team from expectations built on demos.

Be honest about what you know and don't. If leadership hasn't decided how AI affects headcount, say so. "I don't know yet, and here's when I expect to know more" is far less stressful than silence, because silence always gets filled with the worst-case version. And in a market where many people are already weighing whether to japa, silence can cost you good people.

Watch the quiet ones. The developer who goes silent when "AI can do it" comes up is often carrying the most worry. Our culture doesn't always make it easy to say "I'm not okay," so check in. Not in standup. One-on-one.

Quick break: Stakeholder Bingo 🎯

Mark off every phrase you've heard this quarter:

"Can't AI just do it?"	"It's basically a CRUD app"	"How hard can it be?"
"My nephew built this with ChatGPT"	FREE SPACE: "Let's have a quick call"	"Just do a small POC, nothing serious"
"Why is review taking so long?"	"The AI said it works"	"Can we deliver before December?"

Full row? Congratulations. Your prize is knowing you're not alone. 🏆

Where I've landed

So, back to that stakeholder's question: can't AI just do it in a day?

Sometimes, partly. Often, not really. And always with someone who still has to understand it, review it, ship it, and get the call when production breaks on a Sunday.

The fear makes sense. The future is genuinely uncertain, and anyone promising certainty in either direction is selling something (usually a course, sometimes with a "limited slots" flyer).

But a lot of the day-to-day weight of AI anxiety doesn't come from the technology. It comes from how we plan around it, how we talk about it, and whether we're honest with each other while we figure it out.

That part is still up to us.

---

Your turn: how is this showing up on your team? And more importantly, how many did you score on Stakeholder Bingo? Drop it in the comments 👇