Let Chaos Reign, and Then Rein In Chaos

Marc Andreessen argues that companies should give the AI agents their employees build a governed place to work, as they once did for personal computers and spreadsheets.

Marc Andreessen wrote this essay in the newsletter of the venture capital firm a16z on the day Replit launched Replit for Business, a company workspace for AI agents. He argues that useful workplace technology enters as chaos: personal computers, spreadsheets and email all arrived through employees before their companies set rules. The companies that benefited kept that initiative and added just enough control to make the work checkable.

• Employees brought personal computers to work before IT approved them, and some listed them on procurement forms as typewriters or desks.
• Spreadsheets let people who were not programmers model the business, but many were hard to audit; locked cells, version histories and audits made them safe to rely on.
• Email bypassed the company mailroom, and companies answered with permissions and access controls rather than approval for every message.
• He quotes Replit's chief executive, Amjad Masad, that most AI agents now run in chat windows and personal accounts where nobody decided what they could touch.

A team whose staff build AI agents can give those agents a shared place with access rules, version history and an audit trail instead of banning them. That step is an inference. This is an investor's essay published on the day of a product launch it praises, and it reports no measured results.

Derived by Working Surface from the article; more in the Playbook. Source line: You don’t want to put it in a cage. Nor do you want to let it run completely free.

27 October 2026: AI produces reliable work when a team writes down how its own system works and checks the output against something the AI did not write, from frozen rules to audits.