This website uses cookies

Read our Privacy policy and Terms of use for more information.

Prime Ai Solutions

Read time: 4 minutes

Leader, welcome back.

GTA VI scored 99 on Metacritic, tied for the highest ever.
It was built with zero generative AI. Take-Two's shares jumped about 8% the day the reviews landed.

There's a lesson in that for every team working out where AI fits.

LATEST NEWS
The $2bn handcrafted bet just paid off

GTA VI is out, 13 years after GTA V.
Critics gave it 99 on Metacritic, tying Ocarina of Time for the highest score ever.

Was the roughly $2bn development bill worth it? (That's an outside estimate. Rockstar has never confirmed it.)
Pre-orders passed $1.6bn before launch day. Wells Fargo now forecasts 65 million copies sold by March!

On revenue, the bill looks covered within weeks.
Profit takes longer once Sony, Microsoft and the marketing budget take their share. But nobody at Take-Two is losing sleep.

Here's the twist.
Take-Two's CEO, Strauss Zelnick, says generative AI had "zero part" in GTA VI. Every street was handcrafted.

In January, Elon Musk and Epic's Tim Sweeney claimed AI could build a version of GTA VI before Rockstar finished it. It didn't.

But Take-Two isn't anti-AI.
It runs hundreds of AI pilots, and Zelnick says they're already saving cost and time. Humans build what customers pay for.
AI handles the work around it.

My take: most companies have this backwards.
They aim AI at the shiny, customer-facing stuff and leave the back office manual. Flip it.

One more lesson.
Two days before the reviews, Take-Two's shares had dropped 10.5%, and they were down about 15% for the year. Long bets get punished for uncertainty until there's proof.

Your AI project is the same.
Don't promise a transformation. Ship one proof point fast.

Meanwhile, the AI labs went the other way

Rockstar spent 13 years on one product. OpenAI shipped three versions of its Sol model in two months.

On Tuesday it released GPT-6.2 Astra for hard work and a cheaper Sol for everyday tasks.
Google opened Gemini Tasks, which runs jobs in the background, to free users in six countries.
Microsoft launched seven specialised MAI models built to do specific jobs for less.

Early Astra reviews are mixed. Power users say it's hard to steer and sometimes stops short.

My take: rent models like subscriptions you expect to swap.
Spend your real effort deciding which work AI should touch at all.

Which is exactly what this week's prompt does.

STEAL THIS
The Handcraft or Automate audit

Rockstar knew which work was the product and which work was overhead. Most finance teams have never written it down.
This prompt does it in 20 minutes.

First, the input.
Ask everyone on your team to list what they did last week, with rough hours. Ten minutes each. Then paste it in:

ROLE: You are an operations consultant who helps finance teams decide where AI fits and where it doesn't.

ACTION:
1. Sort each task into one bucket:
   - Automate: AI does it, a human checks it
   - Assist: AI drafts, a human decides
   - Handcraft: human only (judgement, relationships, accountability)

2. For Automate and Assist tasks, estimate hours saved per month (as a range) and the main risk.

3. Pick ONE task to pilot first: most hours saved, lowest risk.

4. Design a 2-week pilot for it: the steps, what "good" looks like, and one clear reason to stop.

CONTEXT: [Team size, tools you use (e.g. Excel, Xero, Copilot), and the task list with hours per month.]

EXAMPLES:
- "Matching bank lines to invoices, 12 hrs/month" = Automate. Rules-based and easy to check.
- "First draft of monthly variance commentary, 8 hrs/month" = Assist. AI drafts, the analyst adds the why.
- "Explaining a covenant risk to the bank" = Handcraft. Relationship and accountability.

FORMAT: A table: 
Task | Hours/month | Bucket | Est. hours saved | Main risk. 
Then the pilot plan in under 150 words.

Three tips.

The Examples are the biggest lever. Swap mine for real tasks from your team and the sorting gets much sharper.

Guard your Handcraft column. That's your GTA.
If AI is creeping into it, ask why.

Run it in two different AI tools and compare the pilots they pick.
Where they disagree is where you need a human conversation.

One rule: anonymise anything sensitive before you paste it into a tool your company hasn't approved :)

FROM THE GROUND

This is the work we do at Prime AI: helping businesses build AI into day-to-day operations, not just into a strategy deck.

One example. For one client, we built an AI workflow that finds and qualifies new leads. That led to a new £150k contract.
Across other clients, the same approach has built a pipeline of over £1m!

We're also borrowing a page from HubSpot.
On a16z's podcast this month, co-founder Dharmesh Shah described HubSpot Next, a "startup within a startup". Teams get none of the usual rules. They don't have to build on HubSpot's tech or even use the brand.

Ours is much smaller, but the idea is the same: test new things fast, outside the main business. One that's working is talent.
We've placed several experienced professionals with companies that reached out. If you're hiring, or looking for your next role, hit reply.

Here's the question for your leadership team this week:
What's your GTA, the work only your people can do?
And is AI clearing time for it, or just making noise?

Working through this at your company? It's exactly what we help with.
Hit reply and tell me where you're stuck.

Enjoy the game if you're playing. Some things matter more than month-end.

-Umar Prime AI | primeai.solutions

P.S. Run the audit above, then reply with your top 3 Automate tasks.
I'll tell you which one I'd pilot first.