AI & Tech Developments - Feb 21

Removes AI patterns from text https://github.com/blader/humanizer…

— @tom_doerr

Do you realize just how much fucking money OpenAI and Anthropic have raised? Do you realize how much is at stake? They cannot become commoditized intelligence accessed via API They mathematically and strategically must attack the application layer

— @buccocapital

Cardiologist wins 3rd place at Anthropic's hackathon. Out of 13,000 applications. Built in 7 days by Michał Nedoszytko MD. Coded day and night - in the hospital, in the cloud, while flying from Brussels to San Francisco. A few years ago, it would have been impossible for a doctor

— @trajektoriePL

The exponential continues. Nov 2025: Opus 4.5 had a 5hr 20 time horizon. Feb 2026: Opus 4.6 has a 14hr 30 time horizon. Over three months, that's more than a *doubling* in the duration of coding tasks, measured by how long it takes human professionals, that AI can complete

— @aidigest_

look real hard at the green box every category under 10% on this chart is a department still running on manual effort and tribal knowledge. that's where stress lives that's where budgets live turn one of those workflows into a repeatable AI agent that's vertical saas 2.0

— @gregisenberg

Multi-clauding? Use /rename [label] to name each terminal session

— @_catwu

Anthropic has no strategy. Claude Code started as someone's side project, and so did Cowork and MCP.

— @pmddomingos

I don't get why Anthropic hasn't used Claude Code to rewrite their bloated Electron app as native. Should be much easier than a C compiler.

— @sindresorhus

hey AnthropicAI folks i am paying for Max plan. 37% session used. 7% weekly used. API Error: Rate limit reached dude what am i paying for exactly?

— @sudoingX

.@AlibabaGroup's open source vector database, zvec, went from 500 to 5,300 stars in the past week. It runs inside your app and lets you build local RAG without a separate database. Vector search built-in for local AI tools.

— @greptile

it's WILD that no one has noticed that LLM progress has stalled. All the signs are there

— @DefenderOfBasic

We spin up sandboxes in under 60 milliseconds. For context: you blink at 110ms Built our own orchestrator and stack down to bare metal to help create 1M+ sandboxes per day on our infrastructure. Speed becomes P0 when agents are waiting.

— @ivanburazin

We're going to make $100,000 for this roofing company with @openclaw Heres how any home service business can copy what we're doing

— @_toddanderson

One click, and your data could be exposed to AI agents—forever? Inga Cherney, threat researcher and member of Cato CTRL, discusses how to develop a Zero Trust mindset when working with AI, and how to steer clear of hanging permissions in AI agents.

— @CatoNetworks

The future isn't just imagined — it's engineered. XPENG's Next-Gen IRON embodies our passion for technology, powered by an intelligent core to explore what's next. Follow to explore where it takes you.

— @xiaopenghexpeng

If you pay UGC creators at scale, you know the pain: - Tracking is messy - Paying creators is worse (especially with multiple payout methods) The largest AI tutor app in Europe, Astra AI (2.1M+ users), uses this free tool to track and pay creators with one click

— @lottsnomad

For coding, based on my experience: GLM < Kimi < MiniMax < Gemini < Codex < Claude

— @burkov

has anyone done it? > openclaw > connect to trello > trello has free plan & free api > multiple agents work in organized tasks > you can observe everything / review work good idea or nah?

— @DanKulkov

underrated claude code routine: when you (or Claude) finds issues with your code, have it explain concepts for you dusing your own codebase. here's claude explaining issues my use of useEffect and useCallback. so much nicer than just reading a random doc.

— @pdrmnvd

after using 4.6 a bit i think the chart is showing it's not the best tool for measuring this

— @Icebergy

Fun little app idea: Opportunist. Each night it makes its best guess as to what you'd like and opens a PR on all your projects. Closes them if you don't merge the next day. Notices which things you merge to self-improve.

— @r00k

If Claude Code is capable of fixing security issues in a codebase why can't he write secure code from the start

— @karankendre

i don't think the general public understands what it means that the cost of building any software product is now $200/mo

— @___frye

"99% of products/services still don't have an AI-native CLI yet."

— @steipete

Gemini 3.1 build Figma replacement from 5 hour prompting session. Shipped with Vercel and already at $2400 MRR best time be in the industry

— @krzyzanowskim

Opus 4.6 is *SPECIAL* the same way Sonnet 3.5 was it's doesn't have the highest "base capability" as measured by alpha, but it degrades MUCH slower than other models which is why the time horizons go absolutely ballistic

— @scaling01

Soon: shadcn/skills. Works with shadcn/create, CLI, registries, the full ecosystem. Been testing it for a week. Some really good results.

— @shadcn

bro that mf Dario wasn't kidding about replacing software engineers

— @kanavtwt

Taalas runs Llama 3 8B at 16k tokens per second per user. That's almost an order of magnitude increase even compared to SRAM-based systems like Cerebras. Key idea: each chip is specialized to a given model. The chip is the model. The chat demo is pretty wild:

— @awnihannun

What used to take a whole workday now takes 5-10 mins for Claude Insanity

— @beffjezos

this is the perfect time to have kids cost of the best education is going to go to 0 - ai tutors we are about 10 years from mass produced robots helping us with our house work we are on the precipice of a mass abundance age - value will pivot to producing hardware

— @GRITCULT

Sam Altman: "The inside view at the [frontier labs] of what's going to happen... the world is not prepared. We're going to have extremely capable models soon. It's going to be a faster takeoff than I originally thought."

— @deredleritt3r

if you have AI agents, you need a mission control. here's the exact prompt i used to build mine. paste it into your OpenClaw chat and get a full dashboard in minutes:

— @sharbel

AI is going to eliminate UI like TV eliminated radio

— @shl

Skill Graphs > SKILL .md Everyone's talking about skills for AI agents. But almost nobody is talking about how to structure them. Right now, the default approach is simple. You write one skill file that captures one capability. A skill for summarizing. A skill for code review.

— @akshay_pachaar

I've rebuilt more than 50% of all saas i used to pay for internally and it works 10x better (cuz it's personalized to me and my biz) and costs me zero to run. I have zero doubt that by the end of this year I'll only pay for the things I can't possibly build myself (hosting,

— @johnrushx

It's not because anyone *can* vibe code an app that everyone *wants to*. Anyone can grow their own vegetables, how many do? Automation made it so a single farmer can make 100x more food. Agentic coding makes it so a single developer can make 100x more software.

— @kepano

The future of the software business is solo founders vibe coding 4-5 directly competing SaaS apps that fiercely compete with each other at different price points.

— @typesfast

talked to my friends at microsoft,apple,etc. these mfs said they haven't written any code by hand in months,they dont even look at the code. I am too woke for this and totally against it,but I am also the one unemployed so maybe there is a lesson here

— @DegenSugarBoo

METR evaluation suite is saturated because opus has a 95% CI of up to NINETY EIGHT HOURS

— @seconds_0

Your median decision maker, btw - uses chatgpt to look up stuff - may have tried Claude cowork if they're adventurous - swamped at their job - been told that this ai stuff is gonna solve all our problems - see random posts of ppl supposedly doing cool stuff on LinkedIn /

— @atelicinvest

I just open sourced a full x402 payment gateway. Self-hosted, multi-chain, direct settlement. Fork it, point it at your backend, and start accepting USDC micropayments for any API. No intermediaries holding your funds. Most x402 implementations today rely on hosted

— @ninja_dev3

Stop paying for OpenClaw heartbeats. Got a 16GB Mac Mini? Run a local LLM for heartbeats and save tokens within 10 seconds. Most people waste API credits on keep-alive pings and repetitive low-effort tasks. No need for a $20/mo subscription. A stable local setup runs

— @ziwenxu_

The $100/$200 month plans have given people wildly unrealistic expectations about what this costs today at scale. Most still have no idea how much API spend they'd actually be lighting on fire trying to replicate that in an enterprise setting.

— @HackingLZ

If you can vibe code your idea, you're thinking too small

— @shl

This is true. Anthropic will change something radically soon. They are probably burning money like no other companies before. I have to test other models as well. Minimax 2.5 test was a failure.

— @heyandras

made a little skill for this to share: https://github.com/gbasin/stress-test-skill… i think there's a broader point that... **the models are still lazy and not aware of their capabilities** ...hopefully this is solved in the next gen, but for now you can save a lot of time and pain by having the

— @garybasin

can anyone name a 100 hour software engineering task which is meaningfully a discrete "task" and not a trivial composition of multiple shorter subtasks?

— @tenobrus

Code Mode is all you need, very excited about this direction for MCP https://blog.cloudflare.com/code-mode-mcp/

— @mattzcarey

I did some reverse engineering of OpenClaw's internals today. Insane attention to detail is the secret sauce. There are many very clever (and undocumented) tricks to learn about. Highly recommend.

— @Yampeleg

Steal this idea. Disposable email, phone number and credit cards for AI agents so they can act on your behalf without messing up your real life. I feel this would be inevitable as we give our agents more autonomy on the web.

— @paraschopra

The Cloudflare API has over 2,500 endpoints. Exposing each one as an MCP tool would consume over 2 million tokens. With Code Mode, we collapsed all of it into two tools and roughly 1,000 tokens of context.

— @Cloudflare

Gemini 3.1 Pro does worse than Gemini 3 Pro on Vending-Bench 2.

— @andonlabs

opus 4.6 is literally just smart gpt4o. beware imo

— @bayeslord

I have a dumb question. If AI gets to the point where it can consistently turn $1.00 into $2.00, does that end money as an idea?

— @BoredElonMusk

> installed Antigravity > chose Gemini 3.1 Pro (High) > ask which model it is > telling me it's powered by Claude 3.7 Sonnet Is the UI lying, or is the agent/model lying/hallucinating?

— @Yuchenj_UW

We're all basically context janitors for AI at this point

— @petergyang

This is genuinely insane. The world's most important graph has gone vertical. GPT-5.2 was released Dec. '25. Opus 4.6 in Feb. '26. Doubling time is now 2 months. Down from the initial trend of 7 months. AI solved software. That's the reality we are in now.

— @daniel_mac8

The $200 Claude Max plan seems like a great deal until you realize API pricing for Opus is incredibly expensive and they can't compete in the long run in terms of dollars-per-token-utility at current sticker prices

— @adamdotdev

Lines of code is a terrible productivity metric. So is pull requests per day. But tokens spent really tops it all; not even pretending to measure any outcome, just effort.

— @gunnarmorling

Seeing people say "coding is solved" when I'm fully rewriting a couple weeks of work with Claude Code is a bittersweet thing. I love the tech and it was a good way to explore what needs to be implemented but the reality is that we're still..... so far

— @fbrasisil

pure insanity it's happening METR newest estimate. Claude Opus 4.6 has a 50% time horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks.

— @Dr_Singularity

How did Anthropic ship a desktop app to work with Excel and PowerPoint in an agentic way before Microsoft did?? (FWIW hearing Microsoft has code red because of Claude Cowork - they know they should have shipped something like this first. Still have no response but working on it)

— @GergelyOrosz

If you're looking to buy a Mac Mini, wait 4-6 months, a lot of used Mac Minis in mint condition are about to hit the market

— @fchollet