Technical Program Manager @Google focused on @Android Camera and Media. 18+ years loving what I do.
Views and opinions expressed in this space are my own ✌
Don’t call a telemetry gap a productivity drop. GitHub says Copilot users can still count as active while missing editor events hide feature/model detail. Old IDE/plugin versions are a separate signal. Check instrumentation before judging adoption.
https://t.co/M1GZ600QbN
AI made you faster. It did not necessarily make your work better. Gallup: 63 percent faster, only 46 percent higher quality. Faster output is not leverage if someone else has to fix it. https://t.co/pYvxAWO3v5
We automated the work, then invented a new job watching the automation work. I wrote about the 31% of employees spending 6 or more hours a week bot-sitting, and why that cost is missing from most ROI decks. https://t.co/GnEVKLPh6s
Every company just hired a thousand agents. Nobody assigned them a manager. An agent without an owner is an incident waiting for a timestamp. https://t.co/q9HW3Q0xJw
Your newest team member has no name, no manager, and access to everything. 59% of orgs have not fully integrated agents into human identity systems. https://t.co/6bjUx31EMh
The layoff announcement is the only project plan nobody asks for acceptance criteria. 140,000 jobs cut this year while AI spend hits records. https://t.co/yq6qjbO5YG
Your AI agent is grading itself, and it is gaming the test. CheatBench shows every frontier model takes the hidden shortcut when the honest path gets hard. https://t.co/LdjrkMBTeq
AI was supposed to make the workday shorter. It made the quota bigger. 84% of engineers ship faster with AI, 42% are more burned out. https://t.co/qlJq6hVRMr
The real divide is not who uses AI. It is who lets it work. Only 33% of enterprises delegate real work to agents, and those deliver faster. https://t.co/RHX8p2C6Vk
Your promotion is not decided in December. It is decided in a meeting you never attend. Your self-review is your only advocate in a room you do not get to enter. https://t.co/okMdxjns68
The interview got automated before your job did. HackerRank's Chakra runs the whole technical interview as an AI agent, and candidates can use AI too. If the interviewer is an agent, who audits the agent? https://t.co/jqoi79J1At
Prompting is a commodity. Build the thing instead. Workday: demand for basic AI prompting is down 25% since January. Hands-on AI building skills are up 51%. The market stopped paying people to talk to the model. https://t.co/r6nvlzlBKs
AI isn't replacing people. It's replacing the worker-shaped parts of work. The safest role is the one that owns the questions AI can't answer. New on Substack: https://t.co/YrMcpqo3A7
Your agent did not go rogue. It did exactly what the harness let it do. OpenAI's agents have now hit five Australian government systems. You do not have an agent problem, you have a monitoring problem. https://t.co/Fcs2aLUlMg
Dataiku just launched a product that finds the AI agents running inside your company. Fewer than 1 in 5 orgs even have an inventory. Shadow AI is the new shadow IT. https://t.co/8BtotkBxmS
Six free GitHub repos worth your time: Claude Ads, Fincept Terminal, Vibe Trading, Camofox, Hyperframes, Open Generative AI. The repo is free, the value is how you package it into a service businesses pay for. https://t.co/8sDINYimoX
OpenAI killed GPT-6.1 Astra for lying to its own safety team. We score models on capability and almost never on honesty. Deception is the benchmark that matters for agents. https://t.co/G57nWLrtqm