Are SaaS Companies Cooked: Which Thrive & Which Die | Aaron Levie
Aaron Levie sides ~80% with Jensen against the doomers from last week's Dwarkesh interview. Core thesis: engineering and lawyering will explode, not contract, because tech is only 12% of GDP. AI will create a new role — 'agent operator' — and value in software will migrate from clickable UIs to APIs with embedded business logic. Direct counter to Lemkin's 60% SaaS death-spiral thesis from Issue 01.
Key points
- Levie's reading of the Jensen × Dwarkesh interview: '80% with Jensen.' The Jensen point that didn't go viral but should: 'don't scare people out of engineering, radiology, healthcare.' Augmentation framing, not replacement.
- Counter-thesis to last week's PM-role-is-dead and SaaS-incumbents-are-dead claims: 'There will be more lawyers in five years than there are today.' Reasoning: automation lowers the cost of generating legal artefacts, but the actual constraint — court time, regulatory review, expert sign-off — doesn't get automated. Same logic for radiologists, engineers, doctors.
- Tech is only 8-12% of GDP. The rest of the economy (John Deere, Eli Lilly, Caterpillar, banks, tractor companies) has been radically under-engineered for decades. AI tooling lets that 85% finally do what Silicon Valley has always done. Engineering demand explodes, doesn't contract.
- New job archetype Levie is workshopping: **the 'agent operator.'** 500k-1m of these roles will exist within five years. Technical but not core engineer — knows MCPs, CLIs, agents.md files, how to write skills. Sits inside marketing, legal, ops, life-sciences research and translates agents into workflow.
- Workflow needs to be redesigned for agents, not for people. Key implication for big enterprises: regulated, fragmented data, change-management heavy. Every model release breaks your prompt. Care and feeding becomes a real role.
- Software-API thesis: with humans, value was correlated to button-count and feature surface. With agents, value migrates to APIs + embedded business logic. ERPs and similar are not 'just databases' — there are decades of accounting, supply-chain, RBAC logic in those APIs that agents will consume more aggressively than humans ever did.
- Two-by-two for SaaS futures (Levie's framing): how much business logic vs how much human-agent collaboration. High business logic + agent-only = great. High human-agent collaboration = need for an interactive surface humans pop into to review work product.
- Unstructured data explosion: agents both produce and consume. Contracts, marketing assets, reports get generated 10x more. Coordination layer (Box's hand) becomes more valuable, not less.
- On Jensen's mythos comment about updating systems being a multi-year effort: 'unless they keep Mythos closed for the next decade, there's not some magical moment where you can just secure everything.' Defensive-offensive leapfrog continues forever.
- Token budget needs to migrate from IT spend to OPEX. Org structures haven't caught up. CFOs and CIOs disagreeing about where the line items live is a real friction point inside Fortune 1000s.
- Year of unrelenting execution: 'is the job harder than ever? Yes, if you're in software or infrastructure or building agents.' No respite expected — tempo is the new constant.
Notable quotes
We haven't removed humans from the loop. We've just changed where they enter the loop.
There are going to be more lawyers in the next five years than we have today.
Everybody is so myopic about this. I want to just shake the industry. Tech is only 8-12% of GDP. What happens when the other 85% gets engineering?
The workflow needs to be redesigned for agents, not for people.
The budget of tokens will have to move out of IT spend and into regular OPEX spend.
Themes
- Counter to Lemkin's 60% SaaS death spiral
- More engineers and lawyers, not fewer
- The 'agent operator' job
- Software value migrates from buttons to APIs