Dan Shipper on Riding the Models, Why SaaS Survives, and the Agent Work Surface

Speaker:
Dan Shipper
Source:
Dan Shipper on Riding the Models, Why SaaS Survives, and the Agent Work Surface · ~May 2026
Episode:
Listen →
transcriptlennyaiagentssaasfuture-of-work

Dan Shipper on Riding the Models, Why SaaS Survives, and the Agent Work Surface

Dan Shipper’s return (a year after he called Claude Code early) with a set of deliberately contrarian predictions for how work changes over the coming year — drawn from running Every, a ~30-person fully AI-native company. The episode is to be re-scored in May 2027.

Key ideas

  • Work bifurcates into two agent surfaces. (1) A company “super-agent” you delegate to in Slack (Shopify’s, Ramp’s), trending to one per company because an agent is only useful while a human gardens it; personal agents come later as models get less fiddly. (2) Codex / Claude Cowork as the operating system for your own work — with an in-app browser so the agent sees what you do, and SaaS apps running inside it. “Two agents are better than one.”
  • “The SaaS apocalypse is dumb — I’d buy SaaS stocks.” Agents increase the number of SaaS users, not eliminate them, and because users bring their own tokens (via Codex/Cowork), SaaS margins improve. Build for humans and agents to collaborate on the same artefact: approvals, an inbox/summary, logs, fast rollback, and agent-generated bug reports (which beat human ones).
  • Every agent needs a human. “Automation is a lie”: each automation needs a human keeping it working, and benchmarks overstate autonomy. His “senior-engineer benchmark” (rewrite vibe-coded slop from first principles) jumped to ~62/100 with GPT-5.5 vs a human’s ~90 — but the human reframes the problem (“this is a piece of shit, we must rewrite”), which benchmarks can’t capture. Hence the forward-deployed-engineer role and continued hiring (Every doubled to ~30).
  • “Ride the models” — no job apocalypse. Models “make yesterday’s human competence cheap”, which commoditises it (everyone’s default output looks the same), so humans push ahead using it to make something new. Survival move: use each new model for your work, play, turn the rock over every release. “The edge of AI is wherever AI meets a real human doing something” — not San Francisco.
  • Who thrives: PMs and full-stack designers who get AI-pilled — spiky product/design taste plus light technical skill lets them ship without a team. AI-generated internal writing (plans, email) is fine and often better — provided you stand behind every line. CLIs are “over” (GUIs return); creativity and taste rise in value against the slop.

Topics covered

  • Predicting the future by living in it: Every as a “pocket of the future”; the “reach test”
  • The two work surfaces: company super-agent in Slack; Codex/Cowork as work OS with in-app browser
  • Why SaaS survives and margins improve (users bring their own tokens); building for human+agent
  • The pull-request explosion; deleting to stay coherent; agent-to-agent bug reports
  • “Automation is a lie”; the senior-engineer benchmark; why benchmarks understate remaining human work
  • New roles: the forward-deployed engineer; which roles changed least (CEO, sales)
  • AI-generated internal writing (plans, email) as acceptable and often better
  • Who thrives: AI-pilled PMs and full-stack designers; “ride the models”; find your moment of joy

See also