OpenAI says nearly all its employees have switched from chatbots to Codex agents, but every number comes from OpenAI itself
THE SO WHAT
If 98% of OpenAI employees truly moved from chatbots to Codex agents in under a year, that’s a concrete picture of what agentic AI adoption could look like inside a tech-forward org. For operators, the takeaway isn’t the exact number—it’s to start mapping where task-level agents could replace generic chat in your own workflows and how you’ll measure that shift.
READ THE SOURCE
MORE FROM THE WIRE
Applied AI'Skip the Dyson, get this instead': As a vacuum reviewer, this is the EOFY handstick deal I'd buy myself
The constraint on AI coding platforms is shifting from tokens per second to who signs off on the output. If you’re treating the model as both author and reviewer, you’re building operational risk into your SDLC that won’t show up until an incident.
Applied AIBada bing! The Sopranos is one of the best TV shows ever made, and it's finally coming to 4K Blu-ray
Vendors are now explicitly tying headcount cuts to AI and automation—this is no longer a hypothetical productivity story. If you run a software org, you need a concrete view on which roles shrink, which ones upskill, and how you’ll explain that internally and to the street.
Applied AI‘Never let the builder be its own reviewer’: The next challenge is trust, not speed of code generation
When a White House asks a lab to gate GPT‑5.6 to select partners, model access becomes a policy variable, not just a pricing or capacity decision. If your roadmap assumes open, symmetric access to frontier models, start mapping contingencies around export controls, staged releases, and differentiated partner tiers.
Applied AIElastic announces a ~7% reduction in its workforce, and says "advances in AI and automation are letting us operate with leaner teams"; ESTC closed down 8.70%
A 230M‑parameter model outperforming 4× larger peers on data extraction and running on phones, laptops, and robots is another data point that small, specialized models will eat a lot of enterprise workloads. If you’re defaulting to frontier LLMs for structured data tasks, you’re likely overpaying on latency, cost, and deployment complexity.