Model release pace is relentless in February with Google releasing Gemini-3.1 Pro and AliBaba releasing Qwen3.5-397B-A17B on Chinese New Year's Eve. In terms of easy to use agentic tooling/services this week had Hugo Barra/David Singleton's Dreamer, Howie Lu with HyperAgent and Manus with their AI agents within Telegram chats
Peter Steinberger of OpenClaw joining OpenAI generated buzz as well as Anthropic tightening the squeeze on using their subscriptions outside of their ecosystem
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 15th Feb 2026 to Saturday 21st Feb 2026
1 | via Valerii Anufriev Get insights on building Claude Code pipelines and learn more about Nebius Academy's upcoming 𝐀𝐏 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 London program. Join Denis Volkhonskiy, Rod Rivera... | Stan Fedotov, PhDGet insights on building Claude Code pipelines and learn more about Nebius Academy's upcoming 𝐀𝐏 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 London program. Join Denis Volkhonskiy, Rod Rivera, and me for an in-depth webinar we're hosting on 𝐅𝐞𝐛𝐫𝐮𝐚𝐫𝐲 17. I'm sure the tips and tricks for working with coding agents will pay off both during the course and in your everyday engineering work.  |
2 | I'm a big fan of Replit, I think its a tool people should generally have in their arsenal and most often it helps you get over the cold-start problem of vibe-coding a custom app for your operational workflow as well as building custom apps rather than wrangling an expensive SaaS product. It's different than Cursor, Windsurf, Claude Code, Codex, Droid, Ampcode and many other which are targetted more towards professional developers and even if you swear by these, I encourage you to give the paid plan of Replit a whirl for a couple of months and see what it can do Here's a video of Replit's features it launched in the last year.
.@Replit 25x'ed ARR to >$150m in less than 1 year.
Here's your complete guide to using it in <13 mins: 1:01 - Design Mode 6:26 - Fast v Agent Mode 7:19 - Figma Import 7:48 - Connectors 10:26 - Agent 3 11:25 - Deploy 
|
1 | via Fazri Zubair I couldn't explain this well but I think this TikTok perfectly elaborates my feelings within the last few weeks of being ultra productive but feeling this mental tax from that. |
2 | Peter Steinberger Joins OpenAI as OpenClaw Transitions to Independent Foundation Peter Steinberger, creator of the open-source AI agent project OpenClaw, announced he is joining OpenAI to develop agents and improve Codex. OpenClaw is becoming an open, independent foundation.
I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.
steipete.me
OpenClaw, OpenAI and the future | Peter SteinbergerI'm joining OpenAI to work on bringing agents to everyone. OpenClaw will move to a foundation and stay open and independent. |
1 | Meta's Manus Launches AI Agents in Telegram Chats Manus AI, acquired by Meta for $2 billion in December 2025, rolled out Manus Agents that integrate into Telegram chats for tasks like creating videos, slides, or images from a single message. Users connect tools such as Gmail and Notion, choose between heavy-duty Max or quick Lite modes, and it's free for all subscription tiers with no setup needed. The team promises expansions to WhatsApp, Slack, and more within 30 days. |
1 | From David Singleton (ex-CTO of Stripe) and Hugo Barra (Google/Android -> Xiaomi -> Oculus) comes Dreamer which is billed as A place to discover, build, and enjoy agentic apps. It’s your home for personal intelligence. Currently in beta, signup to be in waitlist (folks pulled in periodically) |
1 | via Yat Siu the fun keeps happening
Perhaps the Metaverse was really meant for Agentic AI exciting to see some of the early Agent Games! simcityagent.com supports .@openclaw @AnimocaMinds @Ethoswarm agents. Its basically an alpha but very fun to see the possibilities looking forward to new developments!
|
1 | Google releases Gemini 3.1 Pro Building on the Gemini 3 series, 3.1 Pro represents a step forward in core reasoning. 3.1 Pro is a smarter, more capable baseline for complex problem-solving. This is reflected in our progress on rigorous benchmarks. On ARC-AGI-2, a benchmark that evaluates a model’s ability to solve entirely new logic patterns, 3.1 Pro achieved a verified score of 77.1%. This is more than double the reasoning performance of 3 Pro.
Gemini 3.1 Pro is here. Hitting 77.1% on ARC-AGI-2, it’s a step forward in core reasoning (more than 2x 3 Pro).
With a more capable baseline, it’s great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative  |
1 | Anthropic launches Claude Code Security, which “scans codebases for security vulnerabilities and suggests targeted software patches” We’re opening a limited research preview of Claude Code Security to Enterprise and Team customers today. Participants will get early access and collaborate directly with our team to hone the tool’s capabilities. We also encourage open-source maintainers to apply for free, expedited access. Making frontier cybersecurity capabilities available to defendersClaude Code Security is one step towards our goal of more secure codebases and a higher security baseline across the industry. 
Opus4.6 found 500+ vulnerabilities in open-source code and we've begun reporting them and contributing patches
quick excerpts from some of them 
Introducing Claude Code Security, now in limited research preview.
It scans codebases for vulnerabilities and suggests targeted software patches for human review, allowing teams to find and fix issues that traditional tools often miss. Learn more: anthropic.com/news/claude-co… 
|
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 Google Nano Banana Pro model within the Freepik tool via the following prompt
a big Persian Villa surrounded by water and nature, village, close view, volumetric lighting, photorealistic, insanely detailed and intricate, Fantasy, epic cinematic shot, trending on ArtStation, mountains, 8k ultra hd, magical, mystical, matte painting, bright sunny day, flowers, massive cliffs, Sweeper3D
Model release pace is relentless in February with Google releasing Gemini-3.1 Pro and AliBaba releasing Qwen3.5-397B-A17B on Chinese New Year's Eve. In terms of easy to use agentic tooling/services this week had Hugo Barra/David Singleton's Dreamer, Howie Lu with HyperAgent and Manus with their AI agents within Telegram chats
Peter Steinberger of OpenClaw joining OpenAI generated buzz as well as Anthropic tightening the squeeze on using their subscriptions outside of their ecosystem
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 15th Feb 2026 to Saturday 21st Feb 2026
1 | via Valerii Anufriev Get insights on building Claude Code pipelines and learn more about Nebius Academy's upcoming 𝐀𝐏 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 London program. Join Denis Volkhonskiy, Rod Rivera... | Stan Fedotov, PhDGet insights on building Claude Code pipelines and learn more about Nebius Academy's upcoming 𝐀𝐏 𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠 London program. Join Denis Volkhonskiy, Rod Rivera, and me for an in-depth webinar we're hosting on 𝐅𝐞𝐛𝐫𝐮𝐚𝐫𝐲 17. I'm sure the tips and tricks for working with coding agents will pay off both during the course and in your everyday engineering work. |
2 | I'm a big fan of Replit, I think its a tool people should generally have in their arsenal and most often it helps you get over the cold-start problem of vibe-coding a custom app for your operational workflow as well as building custom apps rather than wrangling an expensive SaaS product. It's different than Cursor, Windsurf, Claude Code, Codex, Droid, Ampcode and many other which are targetted more towards professional developers and even if you swear by these, I encourage you to give the paid plan of Replit a whirl for a couple of months and see what it can do Here's a video of Replit's features it launched in the last year.
.@Replit 25x'ed ARR to >$150m in less than 1 year.
Here's your complete guide to using it in <13 mins: 1:01 - Design Mode 6:26 - Fast v Agent Mode 7:19 - Figma Import 7:48 - Connectors 10:26 - Agent 3 11:25 - Deploy
|
1 | via Fazri Zubair I couldn't explain this well but I think this TikTok perfectly elaborates my feelings within the last few weeks of being ultra productive but feeling this mental tax from that. |
2 | Peter Steinberger Joins OpenAI as OpenClaw Transitions to Independent Foundation Peter Steinberger, creator of the open-source AI agent project OpenClaw, announced he is joining OpenAI to develop agents and improve Codex. OpenClaw is becoming an open, independent foundation.
I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.
steipete.me
OpenClaw, OpenAI and the future | Peter SteinbergerI'm joining OpenAI to work on bringing agents to everyone. OpenClaw will move to a foundation and stay open and independent. |
1 | Meta's Manus Launches AI Agents in Telegram Chats Manus AI, acquired by Meta for $2 billion in December 2025, rolled out Manus Agents that integrate into Telegram chats for tasks like creating videos, slides, or images from a single message. Users connect tools such as Gmail and Notion, choose between heavy-duty Max or quick Lite modes, and it's free for all subscription tiers with no setup needed. The team promises expansions to WhatsApp, Slack, and more within 30 days. |
1 | From David Singleton (ex-CTO of Stripe) and Hugo Barra (Google/Android -> Xiaomi -> Oculus) comes Dreamer which is billed as A place to discover, build, and enjoy agentic apps. It’s your home for personal intelligence. Currently in beta, signup to be in waitlist (folks pulled in periodically) |
1 | via Yat Siu the fun keeps happening
Perhaps the Metaverse was really meant for Agentic AI exciting to see some of the early Agent Games! simcityagent.com supports .@openclaw @AnimocaMinds @Ethoswarm agents. Its basically an alpha but very fun to see the possibilities looking forward to new developments!
|
1 | Google releases Gemini 3.1 Pro Building on the Gemini 3 series, 3.1 Pro represents a step forward in core reasoning. 3.1 Pro is a smarter, more capable baseline for complex problem-solving. This is reflected in our progress on rigorous benchmarks. On ARC-AGI-2, a benchmark that evaluates a model’s ability to solve entirely new logic patterns, 3.1 Pro achieved a verified score of 77.1%. This is more than double the reasoning performance of 3 Pro.
Gemini 3.1 Pro is here. Hitting 77.1% on ARC-AGI-2, it’s a step forward in core reasoning (more than 2x 3 Pro).
With a more capable baseline, it’s great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative |
1 | Anthropic launches Claude Code Security, which “scans codebases for security vulnerabilities and suggests targeted software patches” We’re opening a limited research preview of Claude Code Security to Enterprise and Team customers today. Participants will get early access and collaborate directly with our team to hone the tool’s capabilities. We also encourage open-source maintainers to apply for free, expedited access. Making frontier cybersecurity capabilities available to defendersClaude Code Security is one step towards our goal of more secure codebases and a higher security baseline across the industry.
Opus4.6 found 500+ vulnerabilities in open-source code and we've begun reporting them and contributing patches
quick excerpts from some of them
Introducing Claude Code Security, now in limited research preview.
It scans codebases for vulnerabilities and suggests targeted software patches for human review, allowing teams to find and fix issues that traditional tools often miss. Learn more: anthropic.com/news/claude-co…
|
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 Google Nano Banana Pro model within the Freepik tool via the following prompt
a big Persian Villa surrounded by water and nature, village, close view, volumetric lighting, photorealistic, insanely detailed and intricate, Fantasy, epic cinematic shot, trending on ArtStation, mountains, 8k ultra hd, magical, mystical, matte painting, bright sunny day, flowers, massive cliffs, Sweeper3D
An X article on what Replit shipped in 2025. I'm expecting an updated version of its Agent very soon
An interesting take on how it pivoted to find its ICP [ideal customer profile]
Here's a lightning course about Replit from a senior Googler on Maven. IMHO, Maven has S-tier teachers
3 | via Yat Siu |
4 | Alibaba Unveils Massive Qwen3.5-397B-A17B AI Model on Chinese New Year's Eve The flagship Qwen3.5-397B-A17B features 397 billion parameters but activates only 17 billion per token using a sparse Mixture-of-Experts design, delivering up to 19x faster speeds than predecessors while cutting costs. It handles a 262,000-token context expandable to over 1 million, supports 201 languages, and tops benchmarks like 87.8 on MMLU-Pro and records on agentic tasks. Released under Apache 2.0, it's available now on Hugging Face, with hosted options on chat.qwen.ai, fueling China's push in efficient, open multimodal AI. |
5 | Whilst the below is not specifically related to AI, I'd like to invoke my privilege as the curator of this group to go somewhat tangential and share something which I think would interest many in this group as well as friends/acquaintainces of members in this in the group —- An upcoming book from the legendary venture capitalist Bill Gurley. It has great social proof in terms of reviews from pre-release readers https://www.amazon.com/dp/0593799666 I hope that you can put BG2Pod on your listening rotation https://www.youtube.com/@Bg2Pod I've shared this Notion page to many people over the years. Hope that it resonates with some of you |
2 | via Shawn Wang aka swyx of Latent Pod/Cognition fame If you’re building agents top place for learning SOTA [state-of-the-art] will be at Ivan Burazin's (CEO of Daytona) Compute Conference on March 9 at the Chase Center in San Francisco |
2 | via Anthropic Claude Sonnet 4.6 is our most capable Sonnet model yet. It’s a full upgrade of the model’s skills across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. Sonnet 4.6 also features a 1M token context window in beta. Pricing remains the same as Sonnet 4.5, starting at $3/$15 per million tokens. |
2 | via Simon Davis Wrote a guide on how to use AI coding tools as the ultimate writing editor. Thought some of you might find it useful. I turned a coding editor into the ultimate writing tool | Simon DavisI've been writing updates, docs, and long-form content in a coding editor for the past 6 months. Not because I'm a developer, but because it's the ultimate writing tool. AI chat windows break with long documents. They hit context limits, introduce copy-paste nightmares, and there's no way to see what changed. |
3 | For what it's worth, some paid newsletter subscriptions that have been strongly recommended to me by AI leaders in Silicon Valley. Paid subscriptions come with various perks including subscriber-only articles, courses and initial subscriptions to various AI tools/services I pay for one of them, debating on the others primarily because at this point I don't know if I can give them the time commitment to be an effective learner. Don't want to bias anybody as to which one I pay for https://www.lennysnewsletter.com/ https://www.lennysnewsletter.com/p/productpass https://creatoreconomy.so/about wrt courses, some of the ones I am aware are and have probably mentioned earlier are those from Santiago, Elvis, Maven |
via Guy Podjarny, former CTO of Akamai currently founder at Tessl , a package manager and lifecycle platform for agent skills and context. Call for Speakers for 𝐀𝐈 𝐍𝐚𝐭𝐢𝐯𝐞 𝐃𝐞𝐯𝐂𝐨𝐧 𝐋𝐨𝐧𝐝𝐨𝐧 (June 1-2) is open! 𝐂𝐅𝐏 𝐜𝐥𝐨𝐬𝐞𝐬 𝐅𝐞𝐛𝐫𝐮𝐚𝐫𝐲 27𝐭𝐡 |
via Ben Jammin This may be the first real AGI |
2 | Anthropic Bans OAuth Tokens from Consumer Plans in Third-Party Tools Anthropic updated its Claude Code docs to prohibit using OAuth tokens from Free, Pro, or Max subscriptions in any third-party products, including its own Agent SDK, requiring API keys instead for app building. The rule targets open-source tools like OpenClaw and Pi Agent, with enforcement risking bans, while a crackdown limits multiple Max accounts per user. |
3 | Evals act as the PRD for AI engineers, defining success criteria and expected behavior for AI products, making them unlaunchable and untrustworthy without proper evaluation frameworks in place. A recent hour-long conversation between Ankit Shukla and Aakash Gupta on the Product Growth podcast which focuses on building up the skill of creating evals for PM's |
2 | Anthropic Launches Automatic Prompt Caching for Claude API The Claude API now automatically caches static parts of prompts—like system instructions or tools—for reuse, cutting costs to just 10% of standard input tokens on cache hits and reducing latency. Developers no longer need manual tweaks; a single cache_control field at the top level detects and reuses the longest matching prefix. Thariq from the Claude Code team shared tips like ordering prompts from static to dynamic and avoiding cache-breakers, quipping 'Cache Rules Everything Around Me' in a nod to hip-hop. |
3 | via Ben Jammin Google Has been on a tear Pomelli’s latest feature update, ‘Photoshoot’ With Photoshoot, you can start from a single image of your product and easily create high quality, customized product shots to elevate your marketing. |
4 | via Munjal Vaidya This makes things more interesting. No more detecting vulnerabilities but training to exploit them realistically. Changing the conversation itself. Introducing EVMbenchOpenAI and Paradigm introduce EVMbench, a benchmark evaluating AI agents' ability to detect, patch, and exploit high-severity smart contract vulnerabilities.  |
5 | via Simon Davis Eleven Labs crossed an important milestone this week, establishing the framework for AI agents to be insured going forward. Shared a bit about it here: AI Liability Unlocked - LinkedIn Carousel | Simon DavisGoldman Sachs priced the AI-driven software market at $780 billion by 2030. Most of it has been stuck. No insurer would touch AI agents. No policy means no sign-off means no scale. ElevenLabs just unlocked that. Here's the breakdown. 👇 |
6 | Details yet to come but it would seem Anthropic's legal team sent requests to Opencode (an opensource cli harness which supports many models) to have them disable use of Anthropic Pro/Max subscriptions and require Opencode users who want to use Opus/Sonnet/Haiku models within that harness to use API keys [and therfore API pricing] |
7 | via Ben Jammin They're specifically allowing Codex CLI use with OpenClaw, whereas Anthropic still is unclear on their stance. |
8 | Some responses based on a question from David James asking
Ben Jammin responding
https://t.me/c/2073191675/2803 Yat Siu responding with it
Also |
2 | From Howie Lu CEO/co-founder of Airtable comes HyperAgent I've been personally burning through billions of tokens a week for the past few months as a builder. Today, I'm excited to announce Hyperagent, by Airtable. Hyperagent gives agents a full computing environment — real browser, shell, filesystem — along with powerful tools like image and video generation, mapping, data warehouse access, and hundreds of enterprise integrations. Think OpenClaw, but secure and cloud-based from the get-go — with skill learning that compounds over time, one-click deployment into Slack, and a command center to manage your entire fleet of agents at scale. AGI is here. Now harness it.Introducing Hyperagent, a new product from the team at Airtable.  |
3 | via Ben Jammin 4 new Google AI updates in the last 3 weeks. 3 of them in the last 2 days. 🎵 Lyria 3 (AI Music Generation) https://blog.google/innovation-and-ai/products/gemini-app/lyria-3/ 👁 Gemini 3 Flash with Agentic Vision https://blog.google/innovation-and-ai/technology/developers-tools/agentic-vision-gemini-3-flash/ 🧠 Gemini 3.1 Pro https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/ Pomelli Photoshoot https://blog.google/innovation-and-ai/models-and-research/google-labs/pomelli-photoshoot/ |
4 | Claude Code Introduces Built-in Git Worktree Support for Parallel Agent Isolation Claude Code added git worktree support to its CLI, Desktop app, IDE extensions. This enables multiple agents to run in parallel within the same repository without interfering with each other. Users can activate it via claude --worktree, the Desktop app's worktree mode checkbox, by instructing subagents to use worktrees, or by adding 'isolation: worktree' to custom agent frontmatter. |
An X article on what Replit shipped in 2025. I'm expecting an updated version of its Agent very soon
An interesting take on how it pivoted to find its ICP [ideal customer profile]
Here's a lightning course about Replit from a senior Googler on Maven. IMHO, Maven has S-tier teachers
3 | via Yat Siu |
4 | Alibaba Unveils Massive Qwen3.5-397B-A17B AI Model on Chinese New Year's Eve The flagship Qwen3.5-397B-A17B features 397 billion parameters but activates only 17 billion per token using a sparse Mixture-of-Experts design, delivering up to 19x faster speeds than predecessors while cutting costs. It handles a 262,000-token context expandable to over 1 million, supports 201 languages, and tops benchmarks like 87.8 on MMLU-Pro and records on agentic tasks. Released under Apache 2.0, it's available now on Hugging Face, with hosted options on chat.qwen.ai, fueling China's push in efficient, open multimodal AI. |
5 | Whilst the below is not specifically related to AI, I'd like to invoke my privilege as the curator of this group to go somewhat tangential and share something which I think would interest many in this group as well as friends/acquaintainces of members in this in the group —- An upcoming book from the legendary venture capitalist Bill Gurley. It has great social proof in terms of reviews from pre-release readers https://www.amazon.com/dp/0593799666 I hope that you can put BG2Pod on your listening rotation https://www.youtube.com/@Bg2Pod I've shared this Notion page to many people over the years. Hope that it resonates with some of you |
2 | via Shawn Wang aka swyx of Latent Pod/Cognition fame If you’re building agents top place for learning SOTA [state-of-the-art] will be at Ivan Burazin's (CEO of Daytona) Compute Conference on March 9 at the Chase Center in San Francisco |
2 | via Anthropic Claude Sonnet 4.6 is our most capable Sonnet model yet. It’s a full upgrade of the model’s skills across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. Sonnet 4.6 also features a 1M token context window in beta. Pricing remains the same as Sonnet 4.5, starting at $3/$15 per million tokens. |
2 | via Simon Davis Wrote a guide on how to use AI coding tools as the ultimate writing editor. Thought some of you might find it useful. I turned a coding editor into the ultimate writing tool | Simon DavisI've been writing updates, docs, and long-form content in a coding editor for the past 6 months. Not because I'm a developer, but because it's the ultimate writing tool. AI chat windows break with long documents. They hit context limits, introduce copy-paste nightmares, and there's no way to see what changed. |
3 | For what it's worth, some paid newsletter subscriptions that have been strongly recommended to me by AI leaders in Silicon Valley. Paid subscriptions come with various perks including subscriber-only articles, courses and initial subscriptions to various AI tools/services I pay for one of them, debating on the others primarily because at this point I don't know if I can give them the time commitment to be an effective learner. Don't want to bias anybody as to which one I pay for https://www.lennysnewsletter.com/ https://www.lennysnewsletter.com/p/productpass https://creatoreconomy.so/about wrt courses, some of the ones I am aware are and have probably mentioned earlier are those from Santiago, Elvis, Maven |
via Guy Podjarny, former CTO of Akamai currently founder at Tessl , a package manager and lifecycle platform for agent skills and context. Call for Speakers for 𝐀𝐈 𝐍𝐚𝐭𝐢𝐯𝐞 𝐃𝐞𝐯𝐂𝐨𝐧 𝐋𝐨𝐧𝐝𝐨𝐧 (June 1-2) is open! 𝐂𝐅𝐏 𝐜𝐥𝐨𝐬𝐞𝐬 𝐅𝐞𝐛𝐫𝐮𝐚𝐫𝐲 27𝐭𝐡 |
via Ben Jammin This may be the first real AGI |
2 | Anthropic Bans OAuth Tokens from Consumer Plans in Third-Party Tools Anthropic updated its Claude Code docs to prohibit using OAuth tokens from Free, Pro, or Max subscriptions in any third-party products, including its own Agent SDK, requiring API keys instead for app building. The rule targets open-source tools like OpenClaw and Pi Agent, with enforcement risking bans, while a crackdown limits multiple Max accounts per user. |
3 | Evals act as the PRD for AI engineers, defining success criteria and expected behavior for AI products, making them unlaunchable and untrustworthy without proper evaluation frameworks in place. A recent hour-long conversation between Ankit Shukla and Aakash Gupta on the Product Growth podcast which focuses on building up the skill of creating evals for PM's |
2 | Anthropic Launches Automatic Prompt Caching for Claude API The Claude API now automatically caches static parts of prompts—like system instructions or tools—for reuse, cutting costs to just 10% of standard input tokens on cache hits and reducing latency. Developers no longer need manual tweaks; a single cache_control field at the top level detects and reuses the longest matching prefix. Thariq from the Claude Code team shared tips like ordering prompts from static to dynamic and avoiding cache-breakers, quipping 'Cache Rules Everything Around Me' in a nod to hip-hop. |
3 | via Ben Jammin Google Has been on a tear Pomelli’s latest feature update, ‘Photoshoot’ With Photoshoot, you can start from a single image of your product and easily create high quality, customized product shots to elevate your marketing. |
4 | via Munjal Vaidya This makes things more interesting. No more detecting vulnerabilities but training to exploit them realistically. Changing the conversation itself. Introducing EVMbenchOpenAI and Paradigm introduce EVMbench, a benchmark evaluating AI agents' ability to detect, patch, and exploit high-severity smart contract vulnerabilities. |
5 | via Simon Davis Eleven Labs crossed an important milestone this week, establishing the framework for AI agents to be insured going forward. Shared a bit about it here: AI Liability Unlocked - LinkedIn Carousel | Simon DavisGoldman Sachs priced the AI-driven software market at $780 billion by 2030. Most of it has been stuck. No insurer would touch AI agents. No policy means no sign-off means no scale. ElevenLabs just unlocked that. Here's the breakdown. 👇 |
6 | Details yet to come but it would seem Anthropic's legal team sent requests to Opencode (an opensource cli harness which supports many models) to have them disable use of Anthropic Pro/Max subscriptions and require Opencode users who want to use Opus/Sonnet/Haiku models within that harness to use API keys [and therfore API pricing] |
7 | via Ben Jammin They're specifically allowing Codex CLI use with OpenClaw, whereas Anthropic still is unclear on their stance. |
8 | Some responses based on a question from David James asking
Ben Jammin responding
https://t.me/c/2073191675/2803 Yat Siu responding with it
Also |
2 | From Howie Lu CEO/co-founder of Airtable comes HyperAgent I've been personally burning through billions of tokens a week for the past few months as a builder. Today, I'm excited to announce Hyperagent, by Airtable. Hyperagent gives agents a full computing environment — real browser, shell, filesystem — along with powerful tools like image and video generation, mapping, data warehouse access, and hundreds of enterprise integrations. Think OpenClaw, but secure and cloud-based from the get-go — with skill learning that compounds over time, one-click deployment into Slack, and a command center to manage your entire fleet of agents at scale. AGI is here. Now harness it.Introducing Hyperagent, a new product from the team at Airtable. |
3 | via Ben Jammin 4 new Google AI updates in the last 3 weeks. 3 of them in the last 2 days. 🎵 Lyria 3 (AI Music Generation) https://blog.google/innovation-and-ai/products/gemini-app/lyria-3/ 👁 Gemini 3 Flash with Agentic Vision https://blog.google/innovation-and-ai/technology/developers-tools/agentic-vision-gemini-3-flash/ 🧠 Gemini 3.1 Pro https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/ Pomelli Photoshoot https://blog.google/innovation-and-ai/models-and-research/google-labs/pomelli-photoshoot/ |
4 | Claude Code Introduces Built-in Git Worktree Support for Parallel Agent Isolation Claude Code added git worktree support to its CLI, Desktop app, IDE extensions. This enables multiple agents to run in parallel within the same repository without interfering with each other. Users can activate it via claude --worktree, the Desktop app's worktree mode checkbox, by instructing subagents to use worktrees, or by adding 'isolation: worktree' to custom agent frontmatter. |
Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.
Native multimodal. Trained for real-world agents.
Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
8.6x–19.0x decoding throughput vs Qwen3-Max
201
Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.Native multimodal. Trained for real-world agents.Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.8.6x–19.0x decoding throughput vs Qwen3-Max201
Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.Native multimodal. Trained for real-world agents.Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.8.6x–19.0x decoding throughput vs Qwen3-Max201
Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.Native multimodal. Trained for real-world agents.Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.8.6x–19.0x decoding throughput vs Qwen3-Max201
Subscribe to This Week in All Things AI
Subscribe to This Week in All Things AI
Support This Week in All Things AI
This Week in All Things AIThis Week in All Things AI - Week 8-2026
Support This Week in All Things AI
This Week in All Things AIThis Week in All Things AI - Week 8-2026
This Week in All Things AIThis Week in All Things AI - Inaugural Edition
Greetings friends, Whilst you may have subscribed to my low-volume newsletter which focused on informing what had changed in the various public Notion pages that I curate, I decided to experiment with a new publication/newsletter which I hope to send on a weekly basis that aggregates what was posted the previous week on the 'All Things AI' telegram group Some of the reasons that I want to do this experimentThere are folks who have mentioned to me that they are not on Telegram and don't wish t...
This Week in All Things AIThis Week in All Things AI - Week 35-2025
Sunday 24th August 2025 and Saturday 30th August 2025
This Week in All Things AIThis Week in All Things AI - Week 39-2025
Sunday 21st September 2025 to Saturday 27th September 2025
This Week in All Things AIThis Week in All Things AI - Inaugural Edition
Greetings friends, Whilst you may have subscribed to my low-volume newsletter which focused on informing what had changed in the various public Notion pages that I curate, I decided to experiment with a new publication/newsletter which I hope to send on a weekly basis that aggregates what was posted the previous week on the 'All Things AI' telegram group Some of the reasons that I want to do this experimentThere are folks who have mentioned to me that they are not on Telegram and don't wish t...
This Week in All Things AIThis Week in All Things AI - Week 35-2025
Sunday 24th August 2025 and Saturday 30th August 2025
This Week in All Things AIThis Week in All Things AI - Week 39-2025
Sunday 21st September 2025 to Saturday 27th September 2025
>800 subscribers
>800 subscribers
Share Dialog
Share Dialog
Long-term memory. Remembers your style, tone, and preferences.
No comments yet