February ends with some more model releases such as Google's Nano Banana 2 and Bytedance's Seedream 5.0 Lite which btw was used to create the header image for this weeks newsletter. Prompt used to generate the header image is at the end of the newsletter
Other notable things releases where Notion Custom Agents, updates to Replit Agent , various additional features to Claude Cowork as well as Wispr Flow's availability on Android along with it offering free, unlimited dictation for ALL users.
Zack Shapiro's tweet thread where he described how his small lawfirm used Claude to massively improve their workflow received a lot of coverage
Readers who think others in their family, friends and acquaintances who are curious in knowing more about rapidly evolving AI tools/services/use cases and would benefit from being subscribers of this weekly newsletter are encouraged to share this publication link to them and invite them to subscribe
The following messages were posted on the 'All Things AI' Telegram group from Sunday 22nd Feb 2026 to Saturday 28th Feb 2026
1 | via Fazri Zubair I’ve been doing a lot of dev this past week, and 5.2 still feels stronger for reasoning across a larger codebase, spotting root causes, catching edge cases, and thinking through test failures. 5.3 is noticeably faster, but it often feels more "task/patch-driven," like it goes straight for the obvious fix instead of stepping back and intuiting where the issue might actually be coming from. Anyone else seeing the same pattern? Wondering if it’s just me or if that’s been the general consensus. Fazri: That makes a lot of sense. I usually use 5.2 Extra High for planning and 5.3 Codex for execution, but honestly 5.2 still feels better for execution too. It is slower, but I get higher quality results with fewer passes and fewer errors, especially when designing and implementing tests. In that area, 5.2 has been noticeably stronger than 5.3 Codex for me. Coop: When you say 5.3 codex, and 5.2 extra high, are you using 5.3 codex extra high as well? Fazri: Yes codex extra high as well Coop: Yeah. Well in that case you’re likely comparing a smaller model to a larger one and even though it is tuned, larger models literally have more brain power :). GLM 4.7 is amazing. Impossible to run on my max. 4.7 flash can. But it didn’t give the same love 4.7 does. I think codex likely falls in the same box. Folks like Kimi dont do flash / codex models maybe for that reason of losing its true effectiveness. Im still a sonnet / opus maxi though, but 5.3 codex has done very well for me on straightforward features, lacking full architecture execution that Opus High gives me. Disgusting bill though which has led me to GLM and Codex for execution these days. |
2 | For those who use Context7 in their coding harness given that its one of the most recommend MCP server, I came across Nia which claims substantial advances over Context7 as well as Exa. I've not yet given it a decent spin. Impressive raise which includes S-tier angels https://app.trynia.ai/sign-up?ref=6WLOTE51 https://www.trynia.ai/blog/nia-vs-context7 |
1 | Google Antigravity users say their paid Google AI accounts were banned after linking Gemini models via OpenClaw; Peter Steinberger says he may “remove support”
We’ve been seeing a massive increase in malicious usage of the Anitgravity backend that has tremendously degraded the quality of service for our users. We needed to find a path to quickly shut off access to these users that are not using the product as intended. We understand
Pretty draconian from Google. Be careful out there if you use Antigravity. I guess I'll remove support.
Even Anthropic pings me and is nice about issues. Google just... bans? news.ycombinator.com/item?id=471158… |
2 | For what it's worth I am a big fan of Wispr Flow as a voice dication tool and I use it extensively on MacOS. It's also available on iOS and Windows And for a limited time, Flow for Android will offer free, unlimited dictation for ALL users. https://play.google.com/store/apps/details?id=com.wispr.flowapp Below is my referral code which gives you one month free for their Pro subscription |
3 | via Piyush Maheswari Thoughts on this : |
4 | via Prashish Rajbhandari Sharing my workflow as a framework Vibe: A Framework for Agentic Software DeliveryOverview Vibe is an agentic software delivery framework. It gives AI coding assistants a structured way to plan, execute, and verify work inside any repository. Instead of typing instructions into a ...  GitHub - prashishh/vibeContribute to prashishh/vibe development by creating an account on GitHub.  |
1 | Building startups around LLM wrappers is getting more risky by the day "I woke up today and Claude killed my startup." "We built an AI agent that automates ad management for you." "Claude and Manus both released connectors for Meta Ads." "So building here feels pointless." |
2 | OpenAI launched WebSockets in the Responses API to enable low-latency, persistent connections for long-running agents with heavy tool calls. This feature avoids resending full context on every turn by streaming incremental inputs, resulting in speed improvements of 20-40% or more, especially in complex workflows. Developers and tools like Cursor, Cline, and Codex reported up to 50% faster performance on tasks involving multiple tool calls.
Introducing WebSockets in the Responses API.
Built for low-latency, long-running agents with heavy tool calls. developers.openai.com/api/docs/guide… 
All OpenAI models in Cursor are now up to 30% faster!
We've upgraded all users to WebSockets with their Responses API.
We tested @OpenAI's new WebSocket connection mode for the Responses API into Cline and the early numbers are wild.
Instead of resending full context every turn, WebSocket mode keeps a persistent connection, sends only incremental inputs. With 5.2 Codex results vs the standard
Introducing WebSockets in the Responses API.
Built for low-latency, long-running agents with heavy tool calls. developers.openai.com/api/docs/guide…
|
4 | various discussions around the allegation from Anthropic that they were victims to distillation attacks by DeepSeek, Moonshot AI, and MiniMax. |
5 | Some discussion around a post from Summer Yue, Safety and alignment at Meta Superintelligence on her post of her Openclaw disregarding her instruction and proceeding to delete her Inbox. Discussions centered whether there was a hidden agenda given Meta's recent acquisition of Manus and its perceived positioning vis-a-vis Openclaw |
6 | Naval Ravikant's latest podcast titled 'On Artificial Intellgence' (also listed in Spotify/Apple as 'A Motorcycle for the Mind' Vibe Coding Is the New Product Management “There’s been a shift—a marked pronouncement in the last year and especially in the last few months—most pronounced by Claude Code, which is a specific model that has a coding engine in it, which is so good that I think now you have vibe coders, which are people who didn’t really code much or hadn’t coded in a long time, who are using essentially English as a programming language—as an input into this code bot—which can do end-to-end coding. |
7 | This YT channel has hosted pods with founders of Chinese labs like Kimi, StepFun, Zhipu etc + convos around chinese ai stack and infra. Whilstt the conversations are in Mandarin , currently Kimi's has subtitles in English. I'm going to run videos from these channels through NotebookLM and see if they can give me English summaries. Those members who are native Mandarin speakers and can understand the content are welcome to share their opinion of the quality of the podcast. Based on the tone and some early output from NotebookLM, I think the podcast host is top notch https://www.youtube.com/playlist?list=PLwAchVoh-4zNwqxe-MvGNQ9crnQYJhJU3 |
1 | Anthropic is adding new Claude Cowork plugins across: HR, design, engineering, ops, and finance, including financial analysis, investment banking, equity research, private equity, and wealth management, plus a “brand voice” plugin from Tribe AI. |
2 | Cloudflare Releases Vinext, a Vite-Based Rewrite of Next.js Built by One Engineer Using AI for $1,100 in Tokens Cloudflare released Vinext, a Vite-based rewrite of Next.js developed in one week by Steve Faulkner using AI agents for $1,100 in tokens. Vinext replicates Next.js functionality, achieving 4x faster builds and 57% smaller bundles while enabling deployment without vendor lock-in. It has been deployed live on government websites, showing improved performance metrics like 396ms FCP. https://blog.cloudflare.com/vinext/
It’s Next.js Liberation Day.
The #1 request we kept hearing: help us run Next fast and secure, without the lock-in and the costs. So we did it. We kept the amazing DX of @nextjs, without the bespoke tooling, built on @vite. We’re working with other providers to make deployment 
blog.cloudflare.com
How we rebuilt Next.js with AI in one weekOne engineer used AI to rebuild Next.js on Vite in a week. vinext builds up to 4x faster, produces 57% smaller bundles, and deploys to Cloudflare Workers with a single command.
To expand on Cloudflare's announcement today on rebuilding Next.js on Vite, this is everything included:
- Reimplemented the Next.js API surface on Vite - Up to 4x faster builds - ~57% smaller bundles - Rebuilt the majority of Next.js in a week using AI - Deployable to  |
3 | Notion rolled out Custom Agents autonomous AI tools that tackle repetitive tasks like routing bugs, drafting updates, and triaging emails—triggered by schedules, Slack, or events. Aimed at Business and Enterprise users, they're free in beta until May 4, then priced at $10 per 1,000 credits, with early tests showing $90-150 monthly for heavy use. Teams like Ramp saved 2,000 hours with a spending tracker agent, while Notion uses them for reports and feedback; some critique the costs as high for lighter workflows
We spent a fair bit of time deliberating how to price @NotionHQ Custom Agents. We eventually landed on usage-based pricing, so you can pay for the work you get done. In this video, I explain our rationale in a bit more detail:
At @NotionHQ, we believe every business deserves powerful & beautiful tools.
AI is the most important technology of this era. It shouldn't only belong to companies that have AI teams or can afford forward deployed engineers. Today we're launching Custom Agents: — The first 
|
4 | via Tom Ho Another way setup your product to allow AI access via MCP server/interface Very important ways for new products to distribute/be discovered next few years
CLIs are super exciting precisely because they are a "legacy" technology, which means AI agents can natively and easily use them, combine them, interact with them via the entire terminal toolkit.
E.g ask your Claude/Codex agent to install this new Polymarket CLI and ask for any 
introducing polymarket cli - the fastest way for ai agents to access prediction markets
built with rust. your agent can query markets, place trades, and pull data - all from the terminal fast, lightweight, no overhead 
|
5 | Love this sentence from Shawn Wang aka swyx on his commentary on Cognition's Devin 2.2 launch. IMHO, worth reading the threads, Scott Wu's article and Shawn's commentary. I'm not a Devin user myself but this is giving me FOMO
Introducing Devin 2.2 – the autonomous agent that can test with computer use, self-verify, and auto-fix its work. Try it for free!
We’ve also overhauled Devin from the ground up: - 3x faster startup - fully redesigned interface - computer use + virtual desktop ...and hundreds 
some illuminati somewhere decided today was Launch Everything Day but just sharing some personal commentary from this as an analyst:
- Scott admits Devin didn’t even have internal PMF at the 2024 launch. took 6 months to get adoption at first enterprise customer. Models werent  |
6 | OpenCode Go is a low cost ($10/month) subscription designed to bring agentic coding to programmers around the world. it provides generous limits and reliable access to the most capable open source models What's Included Kimi K2.5 GLM-5 MiniMax M2.5 The plan is designed primarily for international users, with models hosted in the US, EU, and Singapore for stable global access. Pricing and usage limits may change as we learn from early usage and feedback. |
1 | Combining news about Claude Cowork, Perplexity Computer and Replit Pro and the new Agent modes on Replit in a single post New in Cowork: scheduled tasks. Claude can now complete recurring tasks at specific times automatically: a morning brief, weekly spreadsheet updates, Friday team presentations.
Scheduled tasks means Claude stopped being software you talk to and became software that works while you sleep. That's a category change, not a feature update.
Think about what Cowork already did to Wall Street. Eleven open-source plugins wiped $285 billion off software stocks Introducing Perplexity Computer. Computer unifies every current AI capability into one system. It can research, design, code, deploy, and manage any project end-to-end.
Introducing Perplexity Computer.
Computer unifies every current AI capability into one system. It can research, design, code, deploy, and manage any project end-to-end. 
Perplexity spent three years building an AI search engine at $20/month flat. "Computer" layers a consumption-based product on top of that entire base.
Max subscribers get 10,000 credits per month. Usage-based pricing. Spending caps. Choose which models run your sub-agents. 
claude.ai
ClaudeClaude is Anthropic's AI, built for problem solvers. Tackle complex challenges, analyze data, write code, and think through your hardest work.
Introducing Perplexity Computer.
Computer unifies every current AI capability into one system. It can research, design, code, deploy, and manage any project end-to-end.
Perplexity just dropped Computer!
Think OpenClaw but multi-model, cloud-based, and connected to your entire stack. No local setup, no single model lock-in. It picks the best model for each step on its own. Claude for reasoning, Gemini for research, Grok for speed. Runs
Introducing Perplexity Computer.
Computer unifies every current AI capability into one system. It can research, design, code, deploy, and manage any project end-to-end.
Testing Perplexity Computer
 So far, the flow is basically... you start the agent, let it run, and wait until it finishes the tasks. What I’m seeing in each agent’s execution is that it selects the model depending on the task. In this example, it handled several tasks with 
More Perplexity Computer testing... now diving into the connectors.
Many apps can be linked with just one click, pretty smooth experience. Already connected GitHub, Linear, Gmail, Drive, and Telegram. Curious to try the WhatsApp Business integration next. Still testing, will 
Testing Perplexity Computer
So far, the flow is basically... you start the agent, let it run, and wait until it finishes the tasks. What I’m seeing in each agent’s execution is that it selects the model depending on the task. In this example, it handled several tasks with
Perplexity Computer one-shotted the Terminal worth $30000/yr
Perplexity just became the the first Al company to truly go head-to-head with the Bloomberg Terminal...
Using Perplexity Computer (with no local setup or single LLM limitation), it was able to build me a terminal with real-time data to analyze $NVDA using Perplexity Finance: 
Introducing Replit Pro — $100/mo, built for power users And Core just got better: now $20/mo (price drop) + a new Economy mode that’s 66% more affordable than what you’re paying today. |
2 | via Munjal Vaidya The Code Modernization PlaybookLearn how leading organizations drive down technical debt and accelerate codebase transformation with our Code Modernization Playbook.  Anthropic's new AI tool can write 67-year-old COBOL code, sending 115-year-old IBM's stock tumbling by 13% - IBM stock has worst day in 26 years, down 25% MoM and countingAncient language is still the bread-and-butter of mainframe systems  |
3 | via Ben Jammin A "smaller" AI model is now beating much bigger, older ones from the same company on tough stuff like math, coding, and handling real tasks. The best part: it's completely free to download and run on your own computer.  Introducing the Qwen 3.5 Medium Model SeriesQwen3.5-Flash · Qwen3.5-35B-A3B · Qwen3.5-122B-A10B · Qwen3.5-27B  More intelligence, less compute.• Qwen3.5-35B-A3B now surpasses Qwen3-235B-A22B-2507 and Qwen3-VL-235B-A22B — a reminder that better architecture, data quality,  |
4 | From Guy Podjarny, ex-CTO of Akamai Today, I’m excited to announce that you can evaluate your skills and optimize them on Tessl. This means you can stop debugging agent output and start shipping quality code, faster: |
5 | Below statement about Spellbook was made on a X post within this thread initiated by the logistics guru Ryan Peterson of Flexport who started of by stating
Claude for legal works seems to work just well as Harvey btw
====== Try Spellbook. More customers than Harvey and Legora combined, focused on contracts, fits exactly in a lawyer’s workflow in a way that wouldn’t make sense for Claude. Backed by realtime market data about what is “normal” in contracts in 80 countries. ====== Legal AI Contract Review & Drafting | SpellbookSpellbook is the #1 Legal AI for for transactional lawyers. Using GPT-4o and other large language models to review and suggest language for your contracts, right in Microsoft Word.  FYI, The above tool was vibe-coded via Replit and is now ontrack to have $100M USD annual revenue this year  Currently working at one of those nonexistent companies (we will hit $100M this year):
replit.com
Spellbook — Customer Story - ReplitReplit is an AI-driven software creation platform where everyone can build, share, and ship apps and websites, fast.
Without Replit, inspiration for Spellbook would have perished and the idea would have died in the shower.
 @replit @amasad
 |
1 | Google just launched it's best Image Generation model yet: Nano Banana 2 It's about 50% cheaper than Nano Banana Pro, better quality and faster You can provide multiple inputs and all of them are included with high quality Nano Banana 2: Combining Pro capabilities with lightning-fast speedOur latest image generation model offers advanced world knowledge, production-ready specs, subject consistency and more, all at Flash speed.  |
2 | I will post a single message wrt Anthropic's recent discussions with the US Department of War without any commentary given that I think this is important news about AI.
A statement from Anthropic CEO, Dario Amodei, on our discussions with the Department of War.
anthropic.com
Statement from Dario Amodei on our discussions with the Department of WarA statement from our CEO on national security uses of AI |
3 | Seedream 5.0 Lite Brings Multimodal Reasoning and Precise Control to Professional Image Creation Seedream 5.0 Lite, the latest generation of BytePlus’ image creation model, became available via API on ModelArk. Built for professional and commercial workflows, Seedream 5.0 Lite advances image generation beyond visual synthesis—bringing stronger multimodal reasoning, higher subject consistency, and more controllable outputs that help teams move from concept to production-ready visuals with fewer iterations. Also available via FreePik and Fal
INTRODUCING: Seedream 5.0 Lite — the next generation of AI image creation is here.
Describe the vibe, reference a sketch, throw in a few images and a wild brief — Seedream 5.0 Lite figures out the rest. Built to understand structure, layout, and complex design intent, it 
|
Seedream 5.0 Lite is here on fal at day 0!
Unified multimodal image generation with deep thinking
Built-in online search. Generate real-time news & trends instantly
Precise control over styles, layouts, and details
Thought this Twitter/X thread and the couple of longish articles within the thread from Zack Shapiro might interest group members who are in the legal profession. Feel free to share to your friends, family and acquaintainces if you think this thread would interest them
The answer, increasingly, is @claudeai. That's basically it.
Not Harvey. Not CoCounsel. Not Spellbook. A general-purpose AI that I've taught how I practice law.
Below is my personal website which aggregates links to many of my socials as well as the various content and community that I curate. Feel free to share this link to others who you think may find this content/community useful to them
The cover image of this newsletter via generated via the Seedream 5.0 Lite model within the Freepik tool via the following prompt
modern minimalist architecture, contemporary pavilion design, symmetrical curved structure, panoramic view, architectural photography, night scene, Canon EOS R5, f/2.8, long exposure, tripod mounted, illuminated curved glass pavilion, translucent curved glass walls, seamless glass panels, matte black steel frame roof, urban park setting, bare winter trees, wet reflective pavement, twilight atmosphere, misty dramatic evening, warm interior illumination, ambient city lights, condensation on glass surface, silhouettes of people moving inside, ultra-realistic, architectural visualization, 2K resolution, rich detail