The 12-Hour Week

Three things happened this week that weren't supposed to happen until 2027. They're unrelated — different companies, different domains — but they all point in the same direction: AI capability is outrunning every timeline.

5 predictions updated8 milestones2 companies refreshed

Key Developments

Opus 4.6 can now work autonomously for a full engineering workday

METR's benchmark measures how long an AI can work on a task without human help. Opus 4.6 scored 719 minutes (about 12 hours) at the 50% mark and 70 minutes at 80% reliability — a 2.45x improvement over Opus 4.5 in just three months. Translation: at reduced confidence, AI can now cover a full engineering workday on its own. That was a 2027 target three months ago.

Confirms P-001 ↗ AI models will handle most aspects of software eng...

Counterpoint

METR's benchmark measures task completion in controlled environments, not production codebases with ambiguous requirements and shifting specs. A 12-hour autonomous run in a benchmark tells you about capability — not about whether anyone will actually trust it for a full workday.

Source →

Cursor hits $2B revenue with 12 employees, eyes $50B valuation

$167 million in revenue per employee. That's the number that has Silicon Valley recalculating what a software company can be. Cursor's new Automations feature makes this even stranger — AI agents that trigger on Slack messages, GitHub events, and PagerDuty alerts without anyone typing a prompt. The coding assistant just became an always-on coding colleague.

Counterpoint

Here's the paradox: if AI really automates 75% of coding, why is a coding tool company worth $50B? That price only makes sense if we're in the gold rush phase — everyone buying shovels — not the aftermath.

Source →

$285B wiped from SaaS stocks in a single day

Salesforce dropped 26% in one session. ServiceNow, Workday, and HubSpot fell 8-15%. The trigger: a Morgan Stanley note arguing that AI agents will eliminate 60% of SaaS seat-based revenue within three years. When the market prices in a thesis that fast, it's no longer a prediction — it's a positioning.

Confirms P-035 ↗ AI will displace 92 million jobs but create 170 mi...

Counterpoint

SaaS has been "dying" every cycle — cloud was supposed to kill on-premise, mobile was supposed to kill desktop. Both times, the incumbents absorbed the threat. Salesforce already has Agentforce at $800M ARR. The question isn't whether SaaS companies will build AI — they already are. It's whether that's enough.

Source →