The Question That Should Drive Your 2026 AI Strategy

Anthropic just published fascinating research on how AI is transforming work inside their own company. They surveyed 132 engineers, conducted 53 interviews, and analysed 200,000 Claude Code transcripts.

The headline numbers are impressive: productivity gains jumped from 20% to 50% in one year. Usage doubled. Engineers are tackling increasingly complex tasks with less oversight.

But here's what actually matters for your business.

The Real Insight Isn't the Productivity Gains

It's how they achieved them.

Anthropic's engineers didn't just start using AI for everything. They developed clear criteria for what to delegate:

• Tasks that are easily verifiable ("I can relatively easily sniff-check correctness")

• Work that's well-defined or self-contained ("If a subcomponent is sufficiently decoupled, I'll get Claude to take a stab")

• Projects outside their expertise where validation is manageable

• Repetitive or boring work ("The more excited I am to do the task, the more likely I am to not use Claude")

One engineer put it perfectly: "It's absolutely amazing for everything where validation effort isn't large in comparison to creation effort."

This isn't about having better AI tools. It's about knowing what work to hand off and what to keep.

That's a strategic decision, not a technical one.

What This Means for Smaller Companies

You don't have 132 engineers running experiments on delegation strategies. You don't have months to figure out what works through trial and error.

But you need the same clarity.

Here's the line that jumped out at me: 27% of AI-assisted work at Anthropic consists of tasks that wouldn't have been done at all without AI.

Not done faster, done at all.

• Exploratory work that never makes it into a roadmap

• Quality-of-life improvements that no one has time for

• Small projects that suddenly become big because AI makes them cheap to extend

For smaller companies with limited resources, that 27% isn't just nice to have. It's competitive advantage sitting on the table.

Backend engineers are building UIs they'd normally hand off. Researchers are creating visualisations instead of waiting for a data team. Non-technical employees are debugging code rather than filing tickets.

Everyone's becoming more "full-stack" - capable of work that used to require specialists.

The question isn't whether you're using AI. It's whether you're ready to capture that kind of value.

What This Means for Your 2026 Planning

The real lesson from Anthropic isn't the exact percentages. It's that their teams developed a shared understanding of what to delegate and how to validate it.

Most smaller organisations don't have that luxury. You need to get to a delegation strategy faster, with fewer experiments, and you can't afford to get it wrong.

So here's the key question for 2026 planning:

If you're in planning sessions right now, try this simple exercise:

1. List 10 recurring tasks or projects your team never quite gets to

2. Circle the ones where checking the output would be relatively quick compared to doing the whole task yourself

3. Those circled items are your first candidates for AI delegation in 2026

That conversation will tell you more than any vendor demo.

What I'm Working On

I'm building something to help companies answer this question practically - a way to quickly assess where you stand, where your "missing 27%" probably lives, and what to focus on first.

Not another framework document. Something you can actually use in planning and day-to-day work.

I'll share more in January.

In the meantime, if you want a sounding board on your 2026 AI plans, hit reply and tell me one area where you suspect value is sitting on the table.

The Atomic Builder is written by Faisal Shariff and powered by Atomic Theory Consulting Ltd — helping organisations put AI transformation into practice.