
🥑DevRel | 🧑💻Co-Founder ➝ @Studio1HQ 📤Newsletter ➝ @DailyAI_Insight ⚡️Building ➝ @raahdotdev / https://www.teapawt.org/t.co/IEMldaIjad
Jezero Crater, Mars 🇮🇳
Joined June 2019
- Tweets15.3K
- Following668
- Followers2.2K
- Likes15.7K
Mr. Ånand retweeted
you can now change T3 Code itself by asking its own agent, and the UI updates while you're using it
it's nightly-only so far. a non-dev on stable asked their agent for it anyway, and the agent cloned the source. they're running a fork now
Mr. Ånand retweeted
Founders,
would you rather have:
- 10,000 followers and no revenue
- 500 followers and 20 paying users
- 0 followers but one great distribution partner
Be honest.
Mr. Ånand retweeted
𝗚𝗲𝗺𝗶𝗻𝗶 𝟰 𝗖𝗮𝗿𝗯𝗼𝗻 𝗶𝘀 𝗔𝗿𝗴𝗼𝗻 𝘄𝗶𝘁𝗵 𝗺𝗼𝗿𝗲 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴: 𝗶𝘁 𝘀𝗽𝗲𝗻𝗱𝘀 𝗲𝘅𝘁𝗿𝗮 𝘁𝗼𝗸𝗲𝗻𝘀 𝗲𝘅𝗽𝗹𝗼𝗿𝗶𝗻𝗴 𝗮𝗻𝗱 𝗰𝗵𝗲𝗰𝗸𝗶𝗻𝗴 𝗶𝘁𝘀 𝘄𝗼𝗿𝗸 𝗯𝗲𝗳𝗼𝗿𝗲 𝗶𝘁 𝗮𝗻𝘀𝘄𝗲𝗿𝘀.
Same 1M context and the same built-in tools as Argon. On hard problems it explores more and checks itself, so a turn uses more tokens and takes longer.
The thinking level is the main dial, and medium is the default. Low cuts tokens and latency, but quality can drop on hard tasks. High spends the most and has the slowest first token, and it's meant for hard math, science and the toughest code and agent work.
The catch is max_output_tokens. Thinking and the answer share that cap, so a tight cap can cut the answer off while you still pay for the thinking. If you cap below 32K today, raise it to 32,768, and lower the thinking level instead when you want it cheaper or faster.
Sync calls default to 65,536 output tokens. For longer jobs, set background to true and raise the cap to 131,072 or 262,144. It all lives inside the 1M window, and the API caps output to whatever space your input leaves.
temperature, top_p and top_k are ignored now, so drop them from your configs.
Gemini 4 Carbon is here
it's our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning
Gemini 4 Carbon is now live via the Gemini API and AI Studio for paid API and Google AI Ultra subscribers
try it today: ai.studio/carbon
Mr. Ånand retweeted
as a founder half my day is calls and the other half is forgetting what we decided on them
a ring that writes up the meeting while I walk to the next one. double press, done. big time saver
then I just ask it what we agreed on and who's doing what
press and talk and it types wherever I am. hands tasks to Claude or ChatGPT across my laptop and phone
and good thing is "Interface" only listens when I press it
🤝 Paid partnership
Mr. Ånand retweeted
someone made a coding agent's tool calls look like a game instead of a log
CrommaCode turns every step the agent usually does silently into visuals, with sound. it's native, built on GPUI (the UI framework behind Zed), and the installer comes in under 50MB
Ha, looks like a lot of folks outside China found the CrommaCode videos, so here's a proper English post:
Basically, it's a gamified ADE. It takes your agent's tool calls, and everything else it used to do silently, and turns them into gorgeous visuals. Sound and SFX are a core part of it too. While building it, I'd just sit there watching it all day.
The post that caught your attention mentioned T3 Code. What I meant was that I'm "distilling" some of T3 Code's great patterns so CrommaCode can support more agent providers.
As for performance, don't worry. It's built on GPUI, not a web view. The installer is under 50MB, and every detail is rendered on the GPU. You know Zed, right? It's basically a gamified Zed. 😂
Have fun! More crazy features have already shipped, they're just locked for now. I'm afraid you'd get completely hooked (there's a real central game server coming later).
Most agents only run when someone types a message.
Proactive agents? They run in the background and start their own work when something needs attention
like a new deploy, a metric moving, a schedule hitting.
One of the cleaner ways this actually works is with three kinds of wake-ups:
> You write it in code. Fresh conversation every run.
> The agent sets followups. Continues the same conversation so it remembers what it was already doing.
> The agent sets it own recuring watch. Fresh conversation each time. Cancels itself when it’s no longer needed.
Difference: the agent doesn’t just wait for your prompt or a fixed cron you wrote. It can create and cancel its own runs and do lot more...
1
2
91
Mr. Ånand retweeted
open models with 1M context are kind of wild for exploring codebases.
gave it the whole hono repo, 467 files and 660K tokens, in one prompt to DeepSeek V4.1 Flash 1017
few seconds later it had a map of the framework and a tour of how a request finds its route, every step pointing at real lines.
then i asked a question, a follow-up, and had it write docs for all 13 modules.
✅ basically gitdiagram + deepwiki + codetour + codeviz in one open source tool.
whole session cost $0.63 on @nebiustf
bring your own key. runs in one docker container, works on your local folders and PRs too
Mr. Ånand retweeted
3 repos for when coding agents keep getting your docs wrong:
1. github.com/agentsmd/agents.m…
a README written for agents. build steps, tests and conventions in one file that Codex, Cursor, Jules and Gemini CLI all read
2. github.com/AnswerDotAI/llms-…
a markdown map of your docs at /llms.txt. Chrome's Lighthouse now checks sites for one
3. github.com/upstash/context7
pulls version-specific docs and code examples into the agent's context, as a CLI skill or an MCP server
Mr. Ånand retweeted
Garry Tan's new harness evals hold the model and the tasks fixed and swap only the agent harness: Claude Code, Codex, Hermes and OpenClaw. If your devtool ships a Claude Code skill, one finding matters more than the rankings.
In Claude Code on Opus 5.5, a skill nobody mentioned was opened 0 times in 30 turns. Add one sentence to the prompt, "Use the brain-first skill.", and it was opened in 30 of 30.
Codex on GPT-6.1 Sol picked up an unmentioned skill in 30 of 30 turns. Hermes on Sol opened it 0 times.
Same skill folder. Useful in one harness, invisible in another.
Rules are the other half. In Claude Code, a rule in the system prompt or the startup file reached all 30 turns, against 18 of 30 with no rule.
And opening a skill isn't running it. Told to run gstack's /investigate, /qa, /review or /ship, Opus loaded the skill text every time and ran none of the steps, across 46 sessions.
Mr. Ånand retweeted
Builders,
what is the smallest product change that created the biggest result for you?
A better headline, faster onboarding, pricing change, demo video, one deleted feature, anything.
Mr. Ånand retweeted
Upload a patient spreadsheet to ChatGPT and every phone number, email and medical record number leaves your device with it.
Agent Guard from @Jozu_AI redacts that on your Mac before the request goes out.
Understand with a small demo, a fictional appointment sheet goes into the ChatGPT desktop app. ChatGPT still summarized all the appointment reasons, but every email and phone number came back as [redacted by Jozu].
How the setup works:
→ Policy Gateway sits between the AI client and the model provider
→ The shipped reference policy already redacts phone numbers
→ You add a custom GuardrailPolicy YAML with a regex rule for emails
→ Restart the gateway, then check agentguard logs --gateway to confirm it worked
One catch: Excel files get masked in place, but PDFs and images can't be, so a matching rule may deny the request instead.
The model keeps the useful context and never sees who the patients are.
Full walkthrough here👇
jozu.com/blog/redact-phi-age…
Mr. Ånand retweeted
Postgres didn't speak SQL for its first 8 years.
Here's how a Berkeley research project that got too popular to support turned into PostgreSQL:
Mr. Ånand retweeted
MIT's free Missing Semester 2026 has a full lecture on coding agents.
10 things the Agentic Coding lecture covers:
1/ How a harness works. Tool results go back into the prompt, so the model runs again on every tool call. The core fits in about 200 lines of code.
2/ Privacy. Most setups send a lot of your code and conversation to the cloud by default.
3/ Tests first. Write the tests, check them yourself, then ask the agent to implement.
4/ Feedback loops. Let the agent run the failing check itself so it can keep iterating.
5/ Parallel agents on git worktrees, so two agents don't step on each other's changes.
6/ MCP, to connect agents to tools like your docs workspace.
7/ Context control: clear it for unrelated tasks, rewind instead of piling on corrections, and know when compaction kicks in.
8/ llms.txt, a denser doc for agents than an HTML page.
9/ AGENTS. md or CLAUDE. md loads in full every session. Skills add a layer so the agent only opens what it needs.
10/ Subagents, so web research doesn't flood the main agent's context.
The other 8 lectures cover the shell, Git, debugging and profiling, packaging and code quality. Notes, exercises and YouTube recordings are all free.
missing.csail.mit.edu/2026
Mr. Ånand retweeted
Cursor's Agent SDK turns an agent into a folder:
instructions.md is the prompt
tools/ has one typed action per file
skills/ holds procedures it loads on demand
channels/ is where it wakes up: PRs, CI, Slack, webhooks, cron
The part I'd use first is evals. An eval sends a fixed message and checks which tools ran, and with what input.
When I benchmarked an AI code reviewer, I gave it read-only tools on purpose. With evals, "never touched a write tool" becomes a CI check instead of a hope.
Test what your agent must not do.
Mr. Ånand retweeted
gputaskman is a terminal task manager that shows what's actually sitting in your GPU memory, including how much an LLM really needs once the KV cache is counted.
A 0.6B model at 32k context needs about 3.5 GB of VRAM just for its KV cache. The weights are 604 MB.
Task Manager tells you "15 GB used". This tries to tell you by what: model weights, KV cache, compute buffers, your browser, or the desktop compositor. In one example on a 16 GB RTX 4080 SUPER, the compositor holds 2.18 GB, the LLM process 1.31 GB and the browser 977 MB.
What it shows:
1. VRAM and GPU activity per process
2. Which runtime and model file each process has loaded, like llama.cpp with a Qwen3 GGUF
3. An LLM's memory split into weights, KV cache and compute buffers, plus the KV cost per token so you can see what a longer context will cost before you set it
4. Spill, how much of each process got pushed out into system RAM
5. A --json mode, so you can log it from a script
Windows with NVIDIA or AMD, Linux, WSL and Apple Silicon. No CUDA Toolkit, admin session or API key needed. Free, MIT.
github.com/shanselman/gputas…
Mr. Ånand retweeted
Read the pricing page.
GPT-6.1 Terra, per 1M tokens:
$1 input, $0.05 cached input, $5 output
GPT-5.6 Terra was $2 input, $12 output
The fine print:
Over 272K input tokens, the whole request is billed at 2x input and 1.5x output
1.05M context, 128K max output
No audio or video input, no fine-tuning
Terminal-Bench 4.0 at max effort: 50.3%, up from 21.5%. OpenAI still says Sol and Astra are the better pick for the hardest coding tasks.
Cheaper per token. Check what it costs per task on your own workload.
Mr. Ånand retweeted
Exact lets a coding agent build an app, run it, see it and test it on web, macOS, iOS and Linux, with no person in the loop.
It's the new framework from Expo co-founder Charlie Cheever, written in Rust. The UI is written in Contract, a small declarative language, and the data logic in TypeScript or Rust. It runs natively, with no webview and no JavaScript bridge in the UI, and on the web a ~20 KB runtime drives the real DOM.
It's built on the assumption that much of the code will be written by AI agents.
→ Ten operations, the same on every host: tree, screenshot, tap, type, state, layout, logs, clock, prefer, perf, all through one script
→ The agent owns the clock, so it moves time with clock +60000 or clock settle instead of waiting on animations
→ Tests are scripts of those same operations and run unchanged on any host
→ contract build --json returns diagnostics with stable ids and source ranges
→ The driver refuses to run a build older than its sources and names the rebuild
→ Mistakes come back as refusals that say what to do
→ A whole app fits in a context window
github.com/expo/exact
Mr. Ånand retweeted
𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗛𝘂𝗻𝘁 𝗶𝘀𝗻'𝘁 𝘁𝗵𝗲 𝗼𝗻𝗹𝘆 𝗽𝗹𝗮𝗰𝗲 𝘁𝗼 𝗹𝗮𝘂𝗻𝗰𝗵 𝗮 𝗱𝗲𝘃𝘁𝗼𝗼𝗹
Worth saving before your next launch:
→ Product Hunt
Free to launch. New accounts need posting access first
→ Peerlist Launchpad
Weekly launches with public voting
→ DevHunt
Only dev tools. Listing is free, a launch week costs $49
→ Show HN
For something people can try right away, no signup wall
→ Uneed
Free if you join the line. Pay to pick your own date
→ Microlaunch
Built around 30 days of exposure after you launch
→ Fazier
Daily launches with a leaderboard
→ BetaList
Early adopters browsing new startups. Paid, pre-launch is fine
→ AlternativeTo
List it next to the tools it replaces. Free, reviewed before it goes live



















