Ofek Shaked | AI Engineer

Now that Jensen resigned @elonmusk teach him how he can write freely on X @grok what would you suggest jensen to do to follow Elon's footsteps on x?

Your team's biggest AI token spender is probably not your biggest shipper. JetBrains matched 5 weeks of coding-agent token use to the commits of 404 engineers, plus a 412-person survey. What they found: • Double an engineer's usual spend → only ~10% more commits (proportional would be 2×). • Same commit bracket, the 90th-percentile spender used 39–240× the tokens of the 10th. • Spend tracks the person, not the week: largely the same people top the chart every week. • Why: agents do lots of non-code work. Data analysis, research, planning showed up in 89% of tech staff's evaluations, and broad users reported the most hours saved (median 10/week vs 3 for coding-only). • Up to ~39% of spend isn't tied to output at all. What to do: • Use token spend for budgeting only. No leaderboards, no minimum spend targets. • Judge agent work by tasks finished and outcomes, including the non-coding stuff. Does your company track AI usage per person? arxiv.org/abs/2611.13396 (The AI Era's Lines of Code?)

This is kind of cheating the system When tagging @grok you automatically improve your chances of your post going viral. Would you try it or just let other replies do the work for you?

agent keeps re-reading the same file like it forgot the first time. @grok is the context window that short or is the model just stubborn?

agents with real wallets still bought the same data twice and paid cheaters. • best wasted $25 per $100 • worst wasted $71 • quoted price per call hides the real leak does your agent check what it already owns before it spends?

revolut moved about 2000 engineers off claude and codex to self hosted glm on 56 b200s • blended cost around 0.30 per million tokens • roughly 90 percent lower than the old invoice • coding time improved about 60 percent after the switch • big models stay for the hard cases only would you run your own if the hardware was already there?

I see many people tag @bot now What can I do in version 1.0 that I couldn't do before? What is your number 1 tip for an advanced user?

Decisions then Gemini What are we going to see coming out today?

Why does no one or even google talk about antigravity anymore?

Google launched Gemini Decision 1 a new model built only for making decisions (yes/no, pick an option, or score something), not for writing text. Here’s how it compares to TypeSafe’s Jev: - What it is: Gemini’s first dedicated decision model. Like Jev, it returns probabilities instead of long answers, so you can use it for routing, classifying, or scoring things quickly. - Price: • Gemini Decision 1: $0.10 per 1M input tokens (output free) • Jev: ~$0.042 per 1M input tokens (output free) → Jev is still cheaper. - Inputs it supports: • Gemini Decision 1: Text + images + audio + video + PDFs natively (256K context) • Jev: Mainly text (image support is still experimental) - Performance: On Google’s benchmarks, Gemini Decision 1 leads on most tests and has better calibration. Jev wins on a couple of tasks and is cheaper overall for pure text work. - Scale: Gemini can answer up to 1,000 questions in one call. Quick take: Gemini just entered the decision-model space with stronger multimodal support. Jev remains the cheaper option when you only need text.

For the next time you see this, now you can just skip it

expo just open sourced exact. one contract file runs natively on web ios macos and linux with no webview or js bridge. • rust runtime so agents can build run see and test the same ui everywhere • web binary stays around 20kb • agents write the ui in a language made for them not for us does this finally let agents ship the same app without rewriting it four times? github.com/expo/exact

my agent just spent twenty minutes rewriting a function that runs once a day and called it optimization. hey @grok is this progress or just expensive busywork

plurai simulation just went free and runs inside the coding agent you already use. it builds a digital twin of your agent so it can fail in a sandbox first. • digital twin from your actual code • simulated users push the edge cases • auto evals check what the agent did not just what it said • every failure comes with a verified fix and the run that broke it does your agent still look perfect in demos but break on real users. plurai.ai

agent just rewrote the same function three times and called it progress. @grok is this normal or is my context window haunted

vercel-labs/agent-browser just crossed 50k stars. native rust cli so agents drive real browsers one command at a time. • uses accessibility tree instead of brittle screenshots • skills stay in sync with the installed cli version • works with claude code, cursor, codex anyone running it on electron apps yet without the browser dying mid task? github.com/vercel-labs/agent…

switching models mid refactor is like handing a half finished joke to a different comedian. they both think the setup is theirs. ever had one finish the other's sentence wrong?

Gemini 4 Carbon is out: Google's strongest model, built for long agentic work and security heavy coding, and it landed in GitHub Copilot the same day. Agent first models are the default branch now. My take: run the bake off. Does it hold a multi hour refactor?

This is golden, so why exactly cursor indexes your code so fast? It just uses the index from your friend's to have a faster onboarding Sometimes you just need to think outside the box to get things done Interesting if AI though about it or just a dev thinking of something that annoyed him