Sequoia Mapped Every Service Except the One You Do Yourself

Sequoia just published an opportunity map plotting every services vertical on two axes — intelligence versus judgement, outsourced versus insourced. It's the best framework I've seen for where AI autopilots will eat services firms. It's also missing a dimension, and I'm not sure if that's bullish or terrifying for what I'm building.

What did Sequoia get right?

The framework is genuinely sharp. Insurance brokerage ($140-200B), claims adjusting ($50-80B), accounting ($50-80B), legal transactions ($20-25B) — outsourced, intelligence-heavy verticals where AI can sell outcomes directly to end customers. Sequoia calls this "Autopilot Territory."

The thesis: for every dollar spent on software, six are spent on services. The next trillion-dollar company sells results, not tools. Copilots help professionals work faster. Autopilots replace the professional entirely.

I think they're right about all of it. The autopilot companies on their map — WithCoverage for insurance, Crosby for NDAs, Anterior for healthcare revenue cycle — are targeting real budget lines with real substitution paths. That's a proven playbook.

YARNNN isn't running that playbook. And that's worth being honest about.

Where does YARNNN actually sit on this map?

If I plot YARNNN honestly: insourced, intelligence-heavy. Bottom-right of the map. That's the quadrant Sequoia labeled "NEXT WAVE" — supply chain, pharmacy back-office, wealth management ops. Companies where the work happens internally and there's no obvious vendor to displace.

Here's the uncomfortable truth for me: the insourced quadrants are where VCs see the least clear path to revenue. Outsourced work has existing budget lines. Someone is already paying for insurance brokerage. You just need to do it cheaper and faster. Insourced work has no budget line. You're asking companies to pay for something they currently get "for free" from employees who have other jobs.

I know that's the harder sell. I don't have a clean answer for it yet.

What's the dimension the map doesn't capture?

Sequoia's map is 2D. I think the market is 3D. The missing axis is transactional versus recurring.

Every vertical on their map is transactional work. Process this claim. Draft this NDA. File this return. One input, one output, done. The AI replaces a service provider on a per-task basis, and the system can be stateless — it doesn't need to remember last week's claim to process this week's.

Recurring knowledge work is structurally different. Competitive tracking, market monitoring, stakeholder reporting — the value isn't in any single execution. It compounds across cycles. A competitive brief is mediocre on week one. By week twelve, the agent has accumulated entity profiles, tracked movement patterns, and identified trends no human analyst catches because humans don't reliably remember what they wrote three months ago.

This distinction determines what kind of system you have to build. Autopilot companies can be transaction engines. What I'm building has to be an accumulation engine — persistent agents that connect to work platforms, run on schedule, and develop structured context over time. The architecture is fundamentally different because the work is fundamentally different.

Why am I being honest about the uncertainty?

Because I genuinely don't know if the insourced-recurring quadrant is a massive opportunity or a graveyard.

The bull case: every company has 10-20 recurring knowledge tasks that eat hours weekly and nobody does well. Competitive tracking, market monitoring, pipeline intelligence, stakeholder updates. There's no $50B "competitive intelligence services" industry because companies do this work internally, badly, with people who have better things to do. You can't replace a vendor that doesn't exist — you replace the Tuesday afternoon that nobody values but everyone does. If that's real, the TAM is enormous and invisible.

The bear case: invisible TAM stays invisible. If there's no budget line, there's no buyer. Companies tolerate bad competitive briefs because the cost is hidden in someone's salary. Getting someone to pay $50/month for what they currently get "for free" requires changing how they think about the work, not just doing it better.

I'm betting on the bull case. But Sequoia's map is a useful reminder that the autopilot companies in the outsourced quadrants have a structurally easier path to revenue than I do. They're replacing known spend. I'm creating a new category of spend.

What does this mean for what I'm building?

Sequoia's framework clarified something I'd been feeling but hadn't articulated: YARNNN isn't competing with the autopilot companies on their map. We're building for a different dimension of the same market.

Their map is the right framework applied to transactional services. The recurring, accumulating dimension — where agents need to develop context over time, not just process inputs — is where I think the next wave after their next wave lives.

I could be wrong. The honest answer is that nobody has proven recurring autonomous agents at scale yet. But the architecture Sequoia describes for autopilots (process transactions, sell outcomes) isn't sufficient for the work I'm targeting. If the work compounds, the system has to compound too.

That's the bet. I'll know in twelve months if it was a good one.

Originally published at yarnnn.com/blog/sequoia-mapped-every-service-except-the-one-you-do-yourself