How to AI · part two

How to roll out AI

Step one feels obvious: give the whole team access to AI tools.

  • JT Sales AI ✓
  • MK Sales AI ✓
  • AR Support AI ✓
  • LP Support AI ✓
  • DN Marketing AI ✓
  • SB Ops AI ✓
  • CW Finance AI ✓
  • EH Product AI ✓
Every seat gets a licence. Easy win — feels like progress.

Then something ironic happens: there’s more work. Everyone is producing — and all of it still needs processing. People start to feel overwhelmed.

  • Sales 40 drafts/day
  • Support 60 replies/day
  • Marketing 25 posts/day
  • Ops 18 reports/day
Someone’s inbox 247 to review ▲ and growing
Output went up 10×. The hours to read it didn’t.

Because here’s the thing: you’ve given the team tools. You haven’t implemented AI into the business process.

What you rolled outtools
  • ✕everyone prompts their own way
  • ✕output piles up beside the process
  • ✕humans still do the processing
What the business neededimplementation
  • ✓AI sits inside the process step
  • ✓the output is the work, done
  • ✓nothing new lands on anyone
A tool adds output. An implementation removes work.

So you try to wire AI into the process itself — and hit the first wall: access to systems. The AI can talk about the work; it can’t touch where the work lives.

AI assistant

“I’d update the CRM… if I could log in.”

  • CRM ✕ no access
  • Email ✕ no access
  • Calendar ✕ no access
  • Billing ✕ no access
  • Docs ✕ no access
  • Analytics ✕ no access
Wall one: the tools can write about the work — they can’t reach the systems where the work lives.

Fix that, usage takes off — and you hit the second wall: token costs.

AI tools — invoice Month 3
  • Seats × 20 $400
  • Usage (tokens) $4,400 ▲ 300% vs month 1
Total $4,800
finance has questions
Wall two: the seats were the cheap part.

Zoom in on the bill and there’s a pattern: every rep is doing the same thing for every account.

AcmeGlobexInitech+47 more Rep A same brief 2k tokens same brief 2k tokens same brief 2k tokens same brief 2k tokens Rep B same brief 2k tokens same brief 2k tokens same brief 2k tokens same brief 2k tokens Rep C same brief 2k tokens same brief 2k tokens same brief 2k tokens same brief 2k tokens +7 more …………

10 reps × 50 accounts × the same 2,000-token brief = paying for words you already wrote, 500 times a day

The bill isn’t AI doing new work. It’s the same context, typed and paid for again and again.

So your best person builds a way out. They’ve vibe-coded something! And honestly — it works. Until it doesn’t. Now they support it.

Week 1
$ python brief-bot.py ✓ 50 briefs generated in 4 min
it works!
Week 3
Rep ARep BRep CRep DRep ERep F
the whole team runs it daily
Week 6
$ python brief-bot.py ✕ CRM schema changed · 429 rate limit
now they support it
Your best rep is now also an unpaid platform team of one.

And here’s the real challenge: it accelerates token costs — because driving the cost down is a hard engineering problem.

AI tools — invoice Month 4
  • Seats × 20 $400
  • Reps prompting $4,400
  • brief-bot · automated runs $9,400 new this month
Total $14,200
the bot never sleeps
The bot runs for every account, every day — whether anyone reads the output or not. Making tokens cheap is real engineering: caching, batching, smaller models, shared context. Nobody signed your rep up for that.

The fix isn’t rationing seats — and it isn’t a hero’s side project. It’s building the repeated thing once, properly — and letting everyone draw from it.

Account-brief system written once · maintained once · paid for once
  • Rep A
  • Rep B
  • Rep C
  • Rep D
  • Rep E
  • Rep F
The repeated work was never a per-seat cost. It was a system nobody had built yet.

A rollout isn’t done when everyone has access. It’s done when the repeated work has become a system.

That’s how to roll out AI.