Software might be on its way out as agents shrink the gap between “I want this” and “done.” Subscription apps look shaky. What survives?
The story
*AI tells:*
– Tidy phase ladder.
– Repeated “phase” transitions.
– Jargon stack: “data moat,” “per-seat,” “lock-in.”
– Big claims with thin examples.
– “Survivors?” list lands canned.
*Rewrite*
🟢 The app you’re building might not need to exist soon
I’ve been chewing on an uncomfortable little gremlin.
Software has been the layer between “I want this done” and “done.” Open app. Click buttons. Feed the office machine. Get result. What if that machine is about to get fired?
Copilot, Cursor, and Claude Code are making working software easier to build. The job shifts from writing every piece by hand to saying what you want and checking the robot intern’s homework. Yay. Also sweaty.
Then it gets rough for subscription software. If custom tools get fast and cheap, paying monthly for a plain project tracker, form builder, CRM, or dashboard feels odd. Like renting a house while your garage printer is coughing up houses.
The apps with better odds have something harder to copy: people already there, private data, approvals, laws, trust, weird industry glue. The value isn’t the screens. It’s what sits behind them.
Then come agents: Operator, Perplexity Assistant, Claude Computer Use. You don’t operate them as much as instruct them. Less “click this.” More “handle this.”
Founders may not wanna cuddle with this part. Why open a spreadsheet if an agent can track expenses, flag weird charges, and ping you each month? The app was the gap between intent and outcome. Agents shrink the gap.
I wrote a longer piece, “The Extinction of Software,” if you wanna sit with the squirm. I also built AgentAutoFlow, free on GitHub, as a small bet on this future.
What software survives? Which categories are you suspecting are made of cardboard?
Recent Comments