Factory Just Hit $5 Billion. Marketing Still Runs a Workshop.
Five months. That is how long it took Factory to jump from $1.5 billion to $5 billion.
The AI coding agent startup closed $200 million this week. Sequoia, Khosla, Blackstone, Insight. Adobe, Nvidia, T-Mobile and Palo Alto Networks are already on the platform. Factory is not selling autocomplete. It is selling Droids — agents that take a software job from ticket to test without a human babysitting every line.
While frontier labs spent the weekend asking each other to slow down, the application layer raised another war chest. That is the actual story this week.
What changed in AI this week
Factory raised $200M at a $5B valuation and moved from coding agents to software factories. Their Router already cuts token spend by more than 60%.
Cognition is sitting near a $48B valuation. Capital is not confused about where the work is going.
OpenAI, Anthropic and Google confirmed they have been talking safety standards for weeks. Fine. The builders did not wait for the memo.
The frontier might pause. Production will not.
Marketing and AI: the CRM just grew a brain
Same 48 hours, Salesforce and Nvidia shipped Koa — the first CRM reasoning model for Agentforce.
Not another writing assistant. A model post-trained on 27 years of CRM workflow intelligence, on synthetic data and no customer records, to update opportunities, route cases, and schedule follow-ups. Salesforce says it matches or beats general models on CRM actions with three times fewer errors. Precision on the right action is up 11%. Context recall is 2.1x.
Formula 1, Xero, UChicago Medicine and 1-800Accountant are already in the pilot.
If you run an agency, read that again. The next marketing stack is not “use ChatGPT on the brief.” It is an agent that actually moves work:
Pulls the live pipeline, not last week’s spreadsheet.
Writes the follow-up that matches the stage.
Books the next action without a Monday standup.
Leaves a log you can audit.
We have rebuilt parts of our own workflow around that idea at Dzine Prodigy. Copy is cheap. A system that moves a lead is not.
Do this before Friday
Pick one campaign that currently dies in a Google Doc.
Write it as a factory, not a prompt: brief in, variants out, QA gate, publish. Give the agent a kill switch and a score. If you cannot measure the run, you do not have a factory. You have a chat.
The labs can debate the speed limit. Your clients will pay the team that already shipped.
Comments