AI shrinks jobs where the work is fixed and grows jobs where work never ends
Atlassian co-founder Mike Cannon-Brookes says law and customer support have fixed work, while software has endless work.
20VC with Harry Stebbings · Anthropic's Super Bowl Ad: Who Won & Lost? | Sierra Hits $150M ARR: Is Customer Support Too Crowded?The debate had turned to lawyers. On a podcast panel about the legal AI startup Harvey, the guests were arguing over whether faster software means fewer people, when Mike Cannon-Brookes, the co-founder of software maker Atlassian, offered a test for thinking it through.
His question, in his words: “Is a particular function input or output constrained?”
By input constrained, he argued, he meant jobs where the amount of work is basically fixed. He gave the example of his own in-house lawyers. If AI makes them twice as efficient, he said, they cannot invent twice as many legal problems. Customer service works the same way. If customers ask 100 questions a day, you answer 100 questions, and if you get twice as many customers, you get about 200 questions.
Software building, he argued, works in the opposite way. The to-do list is never finished, so making engineers faster does not remove the need for engineers. It just means they build more software. Cannon-Brookes said “every category that I know of outside of engineering and product is at existential risk of shrinking seats,” and pointed to Workday, the human resources software company, which has said large employers are adding fewer workers and buying fewer employee accounts.
A fellow guest, investor Rory, added a concrete version of the same point. If you automate installing SAP, Salesforce, or NetSuite, the business systems companies use to run their operations, he said, no company buys two copies because installation got cheaper. It just spends less money doing it.
Cannon-Brookes acknowledged one messy middle. Sales and marketing, he said, look limited by the size of the market, but companies can also dream up other offerings and other approaches to selling. And he questioned how law firms would adapt, since they usually bill by the hour, so finishing work faster can cut their own revenue unless they find another way to charge.
Questions people ask
Does this mean software engineers will never lose jobs to AI?
He argues they are safer, not safe forever. Because companies are competing to release updates constantly, faster coding tools lead them to build more features rather than employ fewer engineers. He points to what he sees now: no one is cutting engineering teams while companies outside engineering freeze hiring.
Why did he mention Workday?
Workday sells software for managing hiring, pay, and human resources, and charges partly by how many employee records a customer manages. When large employers slow hiring, they need fewer records, which slows Workday's growth. Cannon-Brookes cites that slowdown as evidence that AI efficiency is already reducing headcount in nonengineering roles.
Why is hourly billing a problem for law firms using AI?
Most firms charge clients for each hour a lawyer works. If AI lets a lawyer do in two hours what took ten, the client expects a lower bill. The firm would need to charge for results or expertise instead, which is a major change in habit, contracts, and expectations.