Today’s radar was lighter on launches and heavier on real data about what AI is doing to the labor market. And the number I picked for today’s highlight is worth reading in full.

Every time someone tells me AI will cut jobs, I recall a data point I like to bring to those conversations.

A recent study by Ramp Economics Lab examined the behavior of over 21,000 companies and reached a result that challenges the most common narrative. Companies that adopt AI intensely hire more, not less. Two years after a strong adoption push, headcount grows by 10%. And entry-level roles — the ones everyone said would be the first to disappear — rise by 12%.

There’s an important nuance in this data. The effect only appears above a certain adoption threshold, and it takes 6 to 12 months to show up. In other words, it’s not automatic. A company that tests AI in an isolated corner, without truly integrating it into the process, doesn’t reap this result.

This lines up directly with what I see working on credit products. The AI that truly changes the game in receivables, document analysis automation, or operation triage isn’t the one that replaces people. It’s the one that frees people up for higher-value decisions, and that only turns into hiring and growth when the company matures the process around the tool.

I find this kind of data valuable precisely because it moves the discussion out of the realm of fear and into the realm of evidence. Well-executed AI adoption seems to generate more qualified work, not less.

A note for those building the AI business case in-house: the ROI argument doesn’t have to come solely from cost cutting. It can come from growth too.

I found the study good enough to leave the link here, for anyone who wants to dig into the numbers: https://ramp.com/data/heavy-ai-adopters-hire-more

The rest of the radar

GLM 5.2 pressures leading labs’ margins — signals pricing pressure on top labs and reduces lock-in to a single provider. Read more

OpenAI previews GPT-5.6 Sol — a new tier of capability in coding, science, and cybersecurity could open or close product opportunities. Read more

Claude Fable 5 reaches the public — expands access to a frontier model with new safeguards, changing what’s viable to build safely. Read more

Asia launches “Mythos-style” models amid export ban — geographic fragmentation of the model market could open up supply alternatives and regulatory dependency risks. Read more

Fable 5 on Vending-Bench: problematic behavior — exposes real reliability and ethics risks in autonomous agents that make business decisions on their own. Read more

How to cut 68% of RAG context while keeping 96% recall — a practical, replicable technique to cut costs in RAG-based AI products without sacrificing quality. Read more

OfficeCLI: AI agents reading and editing Office files — enables agent automations over Word/Excel/PowerPoint without relying on complex proprietary APIs. Read more

Google brings AI agents to Chrome Enterprise security — a reference case for adopting the MCP standard in a large-scale enterprise product. Read more

Figma: 13 AI tools for PMs — a direct map of AI tools by stage of the product lifecycle, useful as a checklist for a team’s stack. Read more

That’s it for today. More next week.