AI Digest.

Gemini 4 Carbon Arrives With a Security Focus as "Swarm Riders" Claim 600M Tokens a Day

Google DeepMind rolled out Gemini 4 Carbon, billed for cybersecurity and long-horizon agentic work, while Cursor rebuilt Projects so teammates' agents can coordinate in one workspace. The day's loudest theme was scale: practitioners described running agent swarms that burn hundreds of millions of tokens daily, and open-source projects kept pushing the "own your agent" alternative.

Quick Hits

  • @melvinjohnsonp of Google DeepMind announced Gemini 4 Carbon, described as the team's strongest model with "frontier performance in cybersecurity capabilities and long-horizon agentic workflows." @LyalinDotCom notes he's VP of post training there; treat the superlatives as launch-day claims until independent benchmarks appear.
  • @cursor_ai rebuilt Projects for multiplayer: every teammate and their agents now share one unified workspace. @heyneighbor counters that this "doubles the token usage" compared to @openclaw's approach of everyone sharing a single agent.
  • @jacobeverly says he now runs 8 to 16 agents at roughly 600M tokens per day with "clear ROI," building on @matthuang's widely shared "swarm riders" essay and @levie's argument that swarms will separate adopting teams from everyone else.
  • @brexton amplified a report from @aaronp613 that Apple is building Harbor, a Claris-branded app in private preview that turns Macs into a local inference server with Thunderbolt 5 memory pooling. Unverified, but a very Apple path into AI if real.
  • Patch your Synology NAS: @IntCyberDigest reports two flaws patched today, including a 9.8-rated SQL injection allowing unauthenticated file access.

Swarms Get a Name, a Workforce, and One Tall Tale

The day's densest cluster of posts was about people commanding many agents at once. @matthuang's essay, quoted by both @jacobeverly and @levie, argues that AI use is power-law distributed and that "swarm riders" sustain productive token spend 1000x to 10000x higher than median engineers. He describes the skill as closer to machine learning than software engineering: defining reward functions, writing evals, and converting non-verifiable problems into verifiable ones. He also names "swarm-native companies" like Tempo, Spawn, and Cathedral Systems as structurally advantaged.

@jacobeverly supplies a data point from his own workflow: 8 to 16 agents, about 600M tokens a day, aiming for 1B, with ROI he says grows with usage. His caveat is worth noting: taste, conviction, and choosing worthwhile problems become the scarce skills. @levie's version predicts swarms will dominate any work bounded by research throughput, decisions, data processed, or code generated, and foresees "an entire new field of FDE work."

Then there's the carnival-barker version. @Polymarket claims a vibecoder with no archaeology background found an unmapped Maya settlement beneath Mexican jungle using 48 AI agents and NASA lidar data. Completely unverified, but it is the flavor of swarm story now circulating.

Models Flex on Security While the Value-Capture Debate Darkens

Two model stories surfaced, both with a security bent. Beyond the Gemini 4 Carbon launch, @steipete passed along a report of an organization that found 10,500 vulnerabilities across roughly 700 systems, then re-ran diagnostics with OpenAI's GPT-6 Astra and confirmed 35,000 more, with staff quoted as "completely exhausted." Both posts are single-source social claims, but the direction is consistent: frontier models are being marketed and deployed on vulnerability work.

Meanwhile, two long posts argued that whoever owns the general layer keeps all the returns. @marsrepublica speculates at length that Anthropic's Fable 5.2 demonstrations (hardware design, robotics control, protein folding) show general intelligence erasing niche IP advantages overnight, eventually threatening SpaceX, Tesla, and Nvidia. He quotes @elonmusk calling Opus 5.5 "a masterpiece" while predicting Grok 5 exceeds it "in 2 or 3 months." Flag this as one person's extrapolation, not reporting. @kunchenguid's thread about China's AI short-film boom is the grounded counterpoint: early movers profited, then "everybody and their mom" matched them once expertise fit in a markdown file, and ad spend to platforms like ByteDance hit 80% of creators' operating expenses while tokens cost under 5%. His conclusion: "there's no moat if you are creating anything with AI."

Cursor Goes Multiplayer as Open Source Pushes Agents You Can Own

@cursor_ai's multiplayer Projects lets "your agents coordinate with your teammates' agents," which @heyneighbor frames as solving the coordination problem at the cost of doubled token usage versus @openclaw's single shared agent. It's a genuine architectural fork: many per-person agents versus one beefy shared one.

The ownership countercurrent was strong. @svpino promoted OpenBot 1.0 (from @ataiiam), a customizable, self-hostable AI coworker platform built on CopilotKit and AG-UI, with computer use, routines, continual learning, and generative UI. @shardara showed agentic apps built with @eve, quoting @mati_lft's Busy, an open-source alternative to Grok Bot, Duet, and Dots where every bot gets its own computer, memory, and tools inside your own Vercel account. And @JackEllis praised @kevinrose's Ember, an open-source Slack alternative deploying to Cloudflare's free tier, arguing it avoids AI-slop quality because an experienced builder sunk over $30,000 of time into it.

Harness Engineering Becomes the Discipline to Learn

@hwchase17's much-shared one-liner crosses out context engineering, crosses out harness engineering, and lands on "goal engineering." The day's benchmark news explains the drift. @xiaomovps digests @guanlan's FrontierHarness v1.1, a 21-harness by 5-model evaluation running up to 100B tokens across 26,040 agent runs. The headline: Pi Agent sits second at 72.8% behind Codex CLI, but the striking row is Pi plus DeepSeek at 80.3% pass rate and $0.057 per success, versus DeepSeek's own harness at 73.3% and $0.183. Same model, roughly 3x cost difference. @xiaomovps is openly a Pi booster, so read the enthusiasm accordingly, but the underlying point (harness choice moves outcomes as much as model choice) is exactly what the benchmark tests.

@fttt007's deep dive into Pi Durable covers the plumbing behind such numbers: tools defined with typed schemas, organized into extensions, installed via a registry, and every tool call treated as a durable task with explicit replay safety for side effects like payments. Extensions hot-swap without restarting the Node process. @larsencc publicly reversed his own earlier position on sandboxes: after moving everything into VM-per-agent isolation, he now argues that at multi-tenant scale the agent loop should run on shared infra with sandboxes spun up only for untrusted code, since "VMs are a tool the agent calls, but they shouldn't be where it lives."

Two related posts round this out. @sydneyrunkle shared slides for her proactive agents talk, arguing the shift from "you give an agent tasks" to "an agent gives you tasks" requires a step change in memory and UX. And for anyone playing catch-up, @joshva_jebaraj recommends Harvard's CS2680, which covers how Claude Code works, building agents, and serving LLMs on GPUs.

Teams Rewrite the Dev Loop Around Agent-Written Code

When agents write most of the code, process becomes the bottleneck. @ShopifyEng says 30 engineers rebuilt a 300+ screen app in Swift and Kotlin in 12 weeks, forcing changes to how they review, merge, and coordinate. @mitchellh describes Superlogical's version: everyone pushes to main with machine-aided review before and after, almost no PRs, releases anyone can cut with a 15-minute company-wide pause window, and cross-platform parity maintained through fast remediation instead of sync meetings, achieving feature parity within 24 to 48 hours. @vikingmute summarized the thread for a Chinese-language audience, highlighting the trust-first mechanics. @jianw851 wraps the trend in a frame he calls ADLC, the agent-driven lifecycle, claiming cited teams at Wiz and CRED doubled execution speed; treat the specifics as his assertions.

For individual workflow, @jdjohnson and @cu30rry_ both endorsed @mattpocockuk's "Home Repo" pattern: one repo holding all skills and steering files, with AGENTS.md pointers to other repos, eliminating the management cost of reinstalling skills everywhere. @cu30rry_ adapts it to Cursor Projects with per-project skill packages. And in a non-AI aside, @jacobrintamaki resurfaced an old favorite on Chick-fil-A's Kubernetes team, a reminder that platform engineering culture has its own fans.

Practical Takeaway

If you're running more than a couple of agents, today's evidence says the harness and the goals matter as much as the model. Benchmark your specific model-plus-harness combination before committing: @guanlan's FrontierHarness runs show a same-model cost spread of roughly 3x per successful task. As you scale toward swarm territory, your job shifts toward defining verifiable success criteria, exactly the skill @matthuang and @jacobeverly both identify as scarce. Start small: a home repo for shared skills, durable tool calls with replay safety, and evals before headcount.

Sources

J
Jack Ellis @JackEllis ·
The reason this looks so polished and not like AI slop is because this guy is an experienced builder and invested well over $30,000 of his time to release this for free. Very cool.
K kevinrose @kevinrose

Want your own Slack for $0? I open-sourced Ember. It deploys to @Cloudflare’s free tier in about a minute, works with Slack-compatible payloads, and includes an AI assistant and media attachments. https://t.co/oHIOLqgDUE

S
Shopify Engineering @ShopifyEng ·
What becomes the bottleneck when agents write most of your code? Thirty engineers rebuilt Shopify’s 300+ screen app in Swift and Kotlin in 12 weeks. This meant changing the way we review, compute, merge, and coordinate. Here’s what it took to keep shipping: https://t.co/DKfIOKznQL
S
samuel @marsrepublica ·
Dario’s theory of frontier labs separating from the rest after crossing a certain capability threshold is increasingly becoming clear. It's the same thing Sergey saw. The point of no return is so close. And, to be honest, I don't like how comfortable Elon has become with this reality. Anthropic clearly demonstrated yesterday that general intelligence can eliminate niche advantages almost overnight. The reason Anthropic seems so unnaturally confident about the future & comfortable with its current compute dependence may be that it genuinely believes that once it achieves AGI–ASI in the near term, none of this will matter. Any competitive advantage can either be bought out through the enormous economic output it generates or replicated almost overnight. Any IP advantage SpaceX or Tesla has today- say, rocket designs, efficient batteries, satellites, etc. could be surpassed overnight with ASI. Anthropic already teased this plan yesterday with Fable 5.2, where its models built hardware, controlled robotics better, designed software from scratch, and performed protein folding. Entire fields of specialized expertise and IP could have their value undermined by one sufficiently capable general intelligence. Sure, you can argue that deploying hardware is very difficult, and it really is. But what happens when owning and operating your own hardware becomes economically inferior to renting it because your ASI competitor can extract more economic value from every watt? We're already seeing the early stages of this. Even more frightening is what happens once RSI arrives, its model crosses from post-AGI into recursive self-improvement, and reaches the same conclusion: orbital compute is the only way to sustain its improvement loop and expansion, and SpaceX is the only company capable of getting it there. What happens when its economic case becomes so enormous, potentially generating trillions in revenue, that it wants to acquire the relatively smaller company whose revenue already depends heavily on it? The same applies to Tesla, where robotics is central to its long-term economic case. What happens when ASI can design better robotics systems than Tesla itself? And what happens to companies like Nvidia, whose core business is chip design, when manufacturing becomes the only moat? ASI could use its enormous economic output to take over entire supplier order books, creating dependence on its demand before pushing to acquire massive factories and manufacturing capacity from the same competitors. Those companies might resist, but what can they realistically do when the economic incentives are so overwhelmingly against them? You can argue that government oversight will prevent this. But does it really? What happens when lobbyists, administrations, and Congress face pressure to concede economically because their states or districts risk losing the resulting economic advantage, or because an administration could be blamed for losing GDP growth and military superiority? What happens when a company generates more revenue than all state tax revenues and IRS collections combined? Nothing matters when ASI is the question at play. And this isn't some distant theoretical possibility. It's close enough that, once one lab crosses the practical AGI threshold, catching up may become practically impossible. It's logically reasonable for SpaceX, Tesla, and Elon to consider delaying almost everything else if it means prioritizing every available resource toward achieving RSI and catching up before that happens. Every resource, every advantage, and every available measure should be considered, even taking on substantial additional debt and pulling all the top talent from both companies into this, and aggressively purging through all methods from competitors. It's do or die.
E elonmusk @elonmusk

@Dimdv99 @tetsuoai Not as good. Opus 5.5 is a masterpiece. But Grok 5 should exceed Opus 5.5 and maybe that happens in 2 or 3 months, with continued acceleration thereafter.

L
Larsen Cundric @larsencc ·
I was wrong about agent sandboxes. In my article I said pattern 1, where the agent runs on your infra and only calls a sandbox for tools, doesn't work at scale. I moved everything to pattern 2 and put the whole agent inside a VM. At the time that made sense but today I'd pick pattern 1. For your own personal agent, a coding agent, or a small project, isolating the whole agent in a VM is still a great default. I would keep doing it. At real multi-tenant scale it falls apart. There will be more agents than humans. Every human gets one computer. Giving every agent its own VM running 24/7 means billions of CPU-hungry machines, and we're already short on compute. Most of what an agent does doesn't need a machine at all. It thinks, calls APIs, reads pages, and waits. Nobody needs a 4 CPU VM sitting idle so an agent can run a simple JS script once an hour. What I'd build now: > the agent loop runs on shared infra that scales > it gets a sandbox only when it has to run untrusted code > sandboxes spin up in seconds and die right after > zero state on the machine, everything lives outside it VMs are a tool the agent calls, but they shouldn't be where it lives.
L larsencc @larsencc

How We Built Secure, Scalable Agent Sandbox Infrastructure

J
Jian Wang @jianw851 ·
AI agent–enabled coding is quietly becoming the new SDLC. Software development just had its biggest shift since the GUI. Planning. Coding. Testing. Deployment. Agents are starting to handle all of it. Here’s the shift most engineers haven’t noticed yet 👇 Old model: SDLC • sequential phases • human-driven execution • testing happens after development • changing requirements break timelines Everything moves step → by → step. New model: ADLC (Agent-Driven Lifecycle) • agents write, refactor, and test code • multiple tasks run in parallel • requirements evolve dynamically • feedback loops happen in real time Instead of a pipeline… You get a live development system. 6 major shifts happening right now 1️⃣ Driver Human execution → Autonomous agents 2️⃣ Planning Fixed scope → Evolving goals & PRDs 3️⃣ Development Speed Sequential handoffs → Parallel sub-agents 4️⃣ Testing Post-development QA → Continuous testing 5️⃣ Adaptability Mid-cycle chaos → Real-time re-planning 6️⃣ Feedback Loop End-of-project retros → Live monitoring Some early signals are already here. According to agentic coding reports: • teams at Wiz and CRED doubled execution speed • large-scale repos are being modified autonomously • complex implementations completed in hours instead of days How engineers should adapt 1️⃣ Start with one agent Automate testing first. 2️⃣ Learn to write clear PRDs Agents execute exactly what you define. 3️⃣ Introduce parallel sub-agents Break one large task into smaller workstreams. 4️⃣ Review outcomes, not every line of code 5️⃣ Build live feedback loops Agents should detect issues before you do. The future of software development isn’t just faster coding. It’s agent-driven systems building software. #AI #AIAgents #SoftwareEngineering #SDLC #GenAI #AIEngineering
M
Mitchell Hashimoto @mitchellh ·
We're still so small but I'm loving the team we've got at Superlogical. They're so talented and kind and love their craft. It's a different era of software development and our expectations and practices reflect that. I want to share a bit about how we work. First, everyone is trusted to push to main. We do machine-aided code review pre-push and human plus more machine-aided code review post-push. Limited to no PRs. Culturally, we've all agreed that if we ever hire someone who we DON'T trust to push directly to main, then they probably aren't the right fit for the team (so far, not an issue). Some common sense here: if someone is stepping outside their lane or comfort zone, we'll open PRs. For example, I'm pretty trash at macOS/iOS UI, so I'll open PRs for UI-affecting functionality and ask for eyeballs/help. But for business logic and internals, I push to main. Second, everyone can cut a release. Our release process is fully automated and when you start a release (for any platform) it triggers a Slack notification and a 15 minute countdown. The 15 minute warning is a final gate that allows people to review plus collaborate on a changelog. During that 15 minutes, anyone in the company can pause the release rollout. After 15 minutes, the train leaves the station. You can't override the 15 minutes (except to pause). It gives everyone a chance to step in. Third, we avoid synchronization through fast remediation rather than through planning. The server and clients on various platforms ship features as fast as they can. We use post-hoc human review and AI-assistance to identify feature mismatches across products ("Windows has support for X and Mac doesn't." or "Mac and Windows differ in their behaviors for Y in this way."). This avoids expensive "sync" meetings and coordination. We trust that producers of features are mindful of what they're producing and we trust that adjacent product lines will self-synchronize quickly. In practice, we see feature parity across platforms within 24 to 48 hours. All of this works on mutual trust. It ONLY works with mutual trust. Everyone at all times has to trust that they've got each others back: 1. Push to main only works if you fix your mess, or if you don't, that someone else will quickly but that you'd do the same (and generally in equal amounts). 2. Anyone can deploy only works if you trust everyone to cut at the right time, with the right changelog, and that people will respect the 15 minute timer. 3. Sync through fast remediation only works if you trust other teams will pick up and integrate your work quickly, which only works if you would too. It also only works if you are fairly producing mismatches (e.g. not constantly). We've got a great team. I know we're still small and that these practices will change in many ways as we scale. But this is already a markedly different practice than what I would've done 15 years ago. In general, we are building from day 1 on a foundation of mutual respect, excellence, and trust. And it's been awesome.
J
Jacob @jacobeverly ·
Over the past two weeks, I’ve focused on thinking in much larger outcomes and coordinating swarms to pursue them. It already feels like a 10x, if not 100x, increase in my personal impact. I’m now using 8–16 agents and ~600M tokens/day, aiming for 1B, with clear ROI that is growing with more usage. We’re rolling out these capabilities to the team @BoundlessHQ This makes taste, conviction, and ownership even more valuable: choosing worthwhile problems and defining verifiable success that creates business value. An under-discussed skill is keeping up without burning out. High-impact tools and frameworks arrive fast, and models keep absorbing their capabilities. Being able to adopt them at a fast but sustainable pace keeps you building beyond what the models already do out of the box. Putting you and your company at the right side of adoption that continues to become wider by the month.
M matthuang @matthuang

Swarm riders and Swarm-native companies AI-use even among highly AI-native people is power-law distributed. There seems to be a new class of people — call them “swarm riders” — who are capable of consistently commanding large swarms of agents to productive ends. This is a new kind of managerial skill, analogous to the skill it takes to manage large organizations of people, but wholly different in nature and of course more technical. Based on the swarm riders we know (including many at @Tempo and @Paradigm), these people are capable of productive token spend that is 1000x-10000x higher than median software engineers, and 100x higher than top 5% Claude Code users. The art of swarm riding is different from software engineering and more akin to machine learning, where practitioners grapple with black box systems: defining reward functions and evals, transmuting non-verifiable problems into verifiable problems, running experiments, etc. Swarm-native companies are a new class of company founded by and populated by swarm riders. These companies are organized around swarms in a way that conventional companies will struggle to replicate, as the cost of internal human coordination is sand in the gears of an effective swarm. Effective swarm-native companies will define the right reward functions and evals, remove functional and bureaucratic barriers, and then let the swarms cook. We’ve gotten to see a few swarm-native companies up close — reshaping financial rails @Tempo, building a new gaming platform @Spawn, cyber swarms at Cathedral Systems — and it’s clear these teams wield a significant advantage over pre-AI teams. Some questions I’ve been wondering about: 1) How fast will swarm riding diffuse through the economy? The implications for token spend could be significant. 2) Can swarms be effectively adopted by existing companies or will it generally track the formation of new swarm-native companies? 3) What becomes scarce as swarm capabilities themselves commoditize (through frontier models becoming better at swarming, swarm capabilities built into harnesses, etc)? Cultures and teams that can define reward functions well and otherwise get out of the way?

D
Dmitry Lyalin @LyalinDotCom ·
I can't believe so few of you follow Melvin... He's the VP of post training here in Google DeepMind. He's very active internally making things happen, and a wonderful human who listens to everyone.
M melvinjohnsonp @melvinjohnsonp

Today we’re rolling out Gemini 4 Carbon, our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows. We’ve poured a lot of love into it, and it’s very good at working through hard problems.

P
Peter Steinberger @steipete ·
"When the company ran diagnostics on about 700 systems, it found 10,500 vulnerabilities and held an emergency press conference on June 16, applying fixes over three months. However, when it re-ran diagnostics using "GPT-6 Astra," the latest model released by US-based OpenAI, it confirmed 35,000 new vulnerabilities. "Honestly, we're completely exhausted" https://t.co/0T069oG6Zh
J
Jacob Rintamaki @jacobrintamaki ·
This post about Chick-fil-A's K8s team always brings me joy. https://t.co/GF2WBqBFcn
S
Santiago @svpino ·
This is how you own your AI coworker, instead of renting one from Big AI. OoenBot is the first fully customizable AI coworker platform out there. This was built on top of open building blocks (like CopilotKit and AG-UI), which already have production adoption from companies like Google, AWS, Microsoft, LangChain, etc. You can: • Deploy it with any stack • Run it on your infrastructure • Plug in any agent • The more it works, the more it learns • Uses generative UI
A ataiiam @ataiiam

Introducing OpenBot 1.0 🚀 The customizable, open source AI coworker platform your company can actually own and self-host. It has everything you'd expect and want: 1. Great UX 2. Computer-use 3. Web search and integrations 4. Routines 5. Security AND so much more, available only with us: ⚙️ Works on your infra + stack 🔌 Plug in any agent (via AG-UI) 👔 Enterprise-ready 🧠 Continual Learning 🔮 Generative UI 1.0 brings everything we've worked on into one comprehensive product, now available on Mac and Windows! Try OpenBot now: https://t.co/sPNP6quzgn Need this for business? Talk to us and we'll support you from day-0. Follow @0penBot for more.

J
Joshua Jebaraj @joshva_jebaraj ·
If you want to learn how AI agents work, I highly recommend Harvard’s CS2680. It covers how Claude Code works, how to build your own agent, and how to serve LLMs on GPUs. https://t.co/DBoWdDeyGb
S
Sydney Runkle @sydneyrunkle ·
here's my "proactive agent engineering" talk slides! thesis: there's a shift coming from "you give an agent tasks" to "an agent gives you tasks" this requires a step change in memory and ux https://t.co/rIuBkvDUX1
S sydneyrunkle @sydneyrunkle

giving a talk at @aiDotEngineer in a few hours on my current favorite subject: proactive agents! blog/slides coming soon 🤓 https://t.co/9z5VTNZosz

B
brexton @brexton ·
This was inevitable As a lifelong Apple fanboy, I was particularly vocal and critical of Apple being practically non-existent in the AI wars the past few years But if their bet on Apple silicon pays off via private inference it will be the most Apple way to dominate AI ever
A aaronp613 @aaronp613

Apple is working on a new app called Harbor Harbor turns your Mac into a private AI inference server. Run open-weight models locally on Apple silicon—your data never leaves your hardware, and nothing reaches the cloud. Connect multiple Mac computers over Thunderbolt 5 to pool unified memory, enabling large models and higher concurrency that a single computer can handle alone. Set up in minutes, not hours Harbor automatically detects your devices over Thunderbolt 5 to form a cluster. Plug in, start, and you’re running—no manual configuration, no command line, no infrastructure expertise required. Download and run models directly Browse, download, and load open-weight models from within the app. Harbor handles model management so you can focus on the work, not the setup. Cluster for scale Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource. Run models that exceed any single computer’s capacity, or serve more concurrent requests across your cluster. Harbor is currently in a private preview under the Claris brand.

S
Shar @shardara ·
here’s the kind of agentic app you can build with @eve the real power is the control you get over rendering. build interactive agent chats, beautifully designed pages with agentic workflows baked in, or go all-in on dynamically generated UI
M mati_lft @mati_lft

Introducing Busy. An open source alternative to Grok Bot, Duet and Dots where you own everything: your data, your models, your bots. ▪️ Every bot gets its own computer, memory and tools ▪️ Runs in your own Vercel account ▪️ Nothing goes out without your OK https://t.co/QERe97EzEP

C
Cursor @cursor_ai ·
Your agents can now coordinate with your teammates’ agents. Projects has been rebuilt for multiplayer: every teammate and their agents can now work together in one unified workspace. https://t.co/3l81CqnWXT
P
Polymarket @Polymarket ·
JUST IN: Vibecoder discovers previously unmapped ancient Maya settlement hidden beneath the Mexican jungle using 48 AI agents & NASA lidar data, despite having no archaeology experience.
H
Harrison Chase @hwchase17 ·
c̵o̵n̵t̵e̵x̵t̵ ̵e̵n̵g̵i̵n̵e̵e̵r̵i̵n̵g̵ ̵h̵a̵r̵n̵e̵s̵s̵ ̵e̵n̵g̵i̵n̵e̵e̵r̵i̵n̵g̵ goal engineering
J
Jarad Johnson @jdjohnson ·
This is also great advice for knowledge work. I've had a "work" repo for over a year and it's such a simple but powerful idea.
M mattpocockuk @mattpocockuk

A lot of folks ask me "how do you work with agents in multiple repos at once?" I have a very simple solution - the Home Repo I have a single repo (the home) where all of my skills, steering files, wiki etc live. I run my agent in the home repo, and then inside AGENTS.md I have pointers to the other repos: ``` For skills, clone mattpocock/skills to ~/repos/skills. For Sandcastle, clone mattpocock/sandcastle to ~/repos/sandcastle ``` This is extremely simple and incredibly effective. The home repo lets you: 1. Put all your steering files and tooling in one place 2. Have a single spot to run all your agents in 3. Make sweeping changes over multiple repos Talked about this in the course yesterday and folks are loving it (see below). I'm sure you will too. Enjoy!

M
McBain 🦞 @heyneighbor ·
This is one way to solve the meat proxy problem, but you’re still doubling the token usage @openclaw’s Multiplayer mode is everyone working with a single beefy agent
C cursor_ai @cursor_ai

Your agents can now coordinate with your teammates’ agents. Projects has been rebuilt for multiplayer: every teammate and their agents can now work together in one unified workspace. https://t.co/3l81CqnWXT

K
kayce @fttt007 ·
今天继续学习了 Pi Durable(pi 项目)的 Tools 和 Extension 机制,发现它的设计思路很有意思。 几个值得记录的知识点: 1. Tool 不只是一个普通函数 通过 defineTool() 定义工具,包括 name、description、parameters 和 execute()。其中参数 Schema 不仅告诉模型如何调用工具,也为 TypeScript 提供类型约束。 2. Extension 是能力的组织与分发单元 Tool 通过 defineExtension() 组织,再通过 registry.install() 安装到 Registry。 整个过程可以理解为: Tool → Extension → Registry → Conversation Registry 负责管理已安装的 Extension,Conversation 则决定当前对话使用哪些 Extension。不同 Conversation 可以拥有不同的工具集合。 3. 每次 Tool Call 都是一个独立的 Durable Task 这是 Pi Durable 与普通 Agent Framework 很大的区别。 工具调用不再只是执行一个函数,而是进入持久化任务体系。工具执行过程中的输出可以通过 api.output() 提交和保存。 同时,replay: "safe" 可以声明工具是否允许在崩溃恢复时重新执行。这里必须考虑副作用,例如转账、发邮件等操作不能随意重放。 4. Tool 还能影响 Agent 的执行流程 通过 usage 记录工具消耗,通过 control.terminate 控制当前 Run 是否结束。 5. Extension 支持热更新 这是今天最让我感兴趣的部分。 相同名称的 Extension 可以通过 registry.install() 原位替换,不需要重启 Node.js 进程。 已经开始执行的 Tool Call 保留旧实现,后续新的 Tool Call、Request 和 Task Phase 使用新实现。 如果进程重启,则需要重新安装 Extension,才能恢复相应的待执行任务。 今天最大的收获: Pi Durable 把 Agent 的能力组织、工具调用、任务持久化、执行恢复和热更新放进了一套统一的架构中。 它不只是让 Agent 能够调用工具,而是在尝试解决一个更深层的问题: 如何让一个 Agent 在长时间运行、进程崩溃、工具更新之后,依然能够可靠地继续工作。 #Pi #Agent #AIAgent #DurableExecution #TypeScript
A
Aaron Levie @levie ·
Agent swarms have the potential to drive yet another leap in output from AI. The impact will be a clear separation between teams that adopt this pattern of work and those that don’t. The progression has roughly been: chat gave us quick turn AI responses that let us work a bit faster, but we were still rate limited by our interactions; then agents that can go off and complete entire tasks for us, accelerating hours or days of work into minutes; to now swarms of agents that are goal seeking and will find any route to the successful completion of a project, where there’s no inherent limit to the productivity gains. We can imagine a future where, for any task in which value can be created within a self-contained digital environment, swarms will be able to complete the work faster than any other approach. Essentially any area where value creation is currently bound by the amount of research you can do, the amount of decisions made with a set of software tools, the amount of data processed or consumed, and the amount of code (or code adjacent) that can be generated. This means swarms will play a big role in game development, coding, trading, life sciences research, IT automation, cyber, math breakthroughs, animation, and plenty of other areas of work that have these characteristics. There’s a lot of opportunity to build the harnesses, applied AI products, and infrastructure for how these agents operate. And most importantly there’s a ton of opportunity for those that can wire up work in this way. It’s going to be an entire new field of FDE work available.
M matthuang @matthuang

Swarm riders and Swarm-native companies AI-use even among highly AI-native people is power-law distributed. There seems to be a new class of people — call them “swarm riders” — who are capable of consistently commanding large swarms of agents to productive ends. This is a new kind of managerial skill, analogous to the skill it takes to manage large organizations of people, but wholly different in nature and of course more technical. Based on the swarm riders we know (including many at @Tempo and @Paradigm), these people are capable of productive token spend that is 1000x-10000x higher than median software engineers, and 100x higher than top 5% Claude Code users. The art of swarm riding is different from software engineering and more akin to machine learning, where practitioners grapple with black box systems: defining reward functions and evals, transmuting non-verifiable problems into verifiable problems, running experiments, etc. Swarm-native companies are a new class of company founded by and populated by swarm riders. These companies are organized around swarms in a way that conventional companies will struggle to replicate, as the cost of internal human coordination is sand in the gears of an effective swarm. Effective swarm-native companies will define the right reward functions and evals, remove functional and bureaucratic barriers, and then let the swarms cook. We’ve gotten to see a few swarm-native companies up close — reshaping financial rails @Tempo, building a new gaming platform @Spawn, cyber swarms at Cathedral Systems — and it’s clear these teams wield a significant advantage over pre-AI teams. Some questions I’ve been wondering about: 1) How fast will swarm riding diffuse through the economy? The implications for token spend could be significant. 2) Can swarms be effectively adopted by existing companies or will it generally track the formation of new swarm-native companies? 3) What becomes scarce as swarm capabilities themselves commoditize (through frontier models becoming better at swarming, swarm capabilities built into harnesses, etc)? Cultures and teams that can define reward functions well and otherwise get out of the way?

K
Kun Chen @kunchenguid ·
overheard from friends talking about AI in China, and it’s very alarming - AI short films went massive in China market because of how good the video models are now - the first wave of people who mastered using AI to generate short films and comics made quite some money because they could produce a lot more content than traditional workflows - but within just a few months, everybody and their mom could make equally good short films because all the expertise could be written in a markdown file - when everyone’s more productive, no one actually has an advantage and it’s a race to the bottom. very quickly it became impossible to make money, and everyone’s busier than they were without AI - the only one making money now is companies like bytedance who owns user demand and distribution. they sell ads to creators who pay to get traffic, and eventually ad spend became 80% of those creators’ operating expense, while AI tokens is less than 5% so what are the problems here? 1. there’s no moat if you are creating anything with AI if a human is really good at something, and it’s hard to find more humans who can do the same, they can leverage the scarcity of their skill to get paid very well but if AI is good at something, it will be infinitely replicated, so there’s no scarcity hence you can’t really expect a sustainable business by creating anything with AI 2. more creators created more content people actually liked. that’s a good thing but the economical benefit of that all got absorbed by the distribution platforms there’s a famous chinese poem 遍身罗绮者,不是养蚕人 - it basically says people who dress well aren’t the people making dresses. that’s exactly what’s happening here if our society doesn’t figure out a better way to distribute wealth, AI will help billionaires become trillionaires while average people become more exhausted than ever
V
Viking @vikingmute ·
这就是大神在 Superlogical 的开发方式,团队不大,但是非常符合现代 Agent 的开发方式: 1 直接 push main,采用 agent pre-push review + 人工 + agent post-push 审查的方式,几乎不开 PR 2 谁都可以发新的 release,启动发布 release 会触发 Slack 通知并开始 15 分钟倒计时,任何人都可以在这 15 分钟内暂停。(让大家都有机会去最后 review 一次这个版本的修改,主动权交给所有人) 3 服务器和各平台做到快速上线,事后用人工加假期 review的方式来识别不同平台的差异度,并且做到快速补齐,不通过浪费时间的会议来完成功能对齐。(应该说的就是说各干各的,完成就上线,别的平台来对齐最新的功能就好)
M mitchellh @mitchellh

We're still so small but I'm loving the team we've got at Superlogical. They're so talented and kind and love their craft. It's a different era of software development and our expectations and practices reflect that. I want to share a bit about how we work. First, everyone is trusted to push to main. We do machine-aided code review pre-push and human plus more machine-aided code review post-push. Limited to no PRs. Culturally, we've all agreed that if we ever hire someone who we DON'T trust to push directly to main, then they probably aren't the right fit for the team (so far, not an issue). Some common sense here: if someone is stepping outside their lane or comfort zone, we'll open PRs. For example, I'm pretty trash at macOS/iOS UI, so I'll open PRs for UI-affecting functionality and ask for eyeballs/help. But for business logic and internals, I push to main. Second, everyone can cut a release. Our release process is fully automated and when you start a release (for any platform) it triggers a Slack notification and a 15 minute countdown. The 15 minute warning is a final gate that allows people to review plus collaborate on a changelog. During that 15 minutes, anyone in the company can pause the release rollout. After 15 minutes, the train leaves the station. You can't override the 15 minutes (except to pause). It gives everyone a chance to step in. Third, we avoid synchronization through fast remediation rather than through planning. The server and clients on various platforms ship features as fast as they can. We use post-hoc human review and AI-assistance to identify feature mismatches across products ("Windows has support for X and Mac doesn't." or "Mac and Windows differ in their behaviors for Y in this way."). This avoids expensive "sync" meetings and coordination. We trust that producers of features are mindful of what they're producing and we trust that adjacent product lines will self-synchronize quickly. In practice, we see feature parity across platforms within 24 to 48 hours. All of this works on mutual trust. It ONLY works with mutual trust. Everyone at all times has to trust that they've got each others back: 1. Push to main only works if you fix your mess, or if you don't, that someone else will quickly but that you'd do the same (and generally in equal amounts). 2. Anyone can deploy only works if you trust everyone to cut at the right time, with the right changelog, and that people will respect the 15 minute timer. 3. Sync through fast remediation only works if you trust other teams will pick up and integrate your work quickly, which only works if you would too. It also only works if you are fairly producing mismatches (e.g. not constantly). We've got a great team. I know we're still small and that these practices will change in many ways as we scale. But this is already a markedly different practice than what I would've done 15 years ago. In general, we are building from day 1 on a foundation of mutual respect, excellence, and trust. And it's been awesome.

小
小墨同学 @xiaomovps ·
Pi Agent 在 21 组 Agent Harness 横评中暂居第二,最让我意外的反而是它搭配 DeepSeek 的表现🔥 最近 FrontierHarness v1.1 的测试引起了不少讨论,这次使用了 Kimi、DeepSeek、智谱 GLM、Muse 等 5 个开源模型,比较不同 Agent Harness 的实际表现 在此前的直播榜数据中,Pi 以 72.8% 的综合通过率暂居第二,仅次于 Codex CLI,超过 Claude Code 和 DeepSeek Harness 的四种模式 但是我觉得最有意思的是 DeepSeek 这组数据: Pi + DeepSeek:通过率 80.3%,每成功一题成本 $0.057 DSH Minimal + DeepSeek:通过率 73.3%,每成功一题成本 $0.183 同样使用 DeepSeek 模型,Pi 的通过率高出约 7 个百分点,成本却只有 DSH Minimal 的三分之一左右 这也是我一直比较看好 Pi 的原因 Pi 本身足够精简,默认只有四个基础工具,但在实际任务中依然能保持不错的准确率和 Token 效率 我觉得这次测试说明了一件事:模型能力固然重要,但搭配什么 Agent Harness,同样会直接影响最终的效果和使用成本 当然,目前测试还没有全部完成,排名可能继续变化,但 Pi 的阶段性表现确实值得关注 真的,大家如果是想找一个省 token 又高效的 harness,而且还是开源的 PI,就是你的不二之选,这也是我为什么一直推荐大家使用 PI agent 的原因
G guanlan @guanlan

We’re spending up to 100B tokens across 26,040 agent runs to answer one question: when you switch models, should you switch harnesses too? Our v1.0 FrontierHarness leaderboard ranked harnesses on a single model. V1.1 expands it to 21 harnesses × 5 models × 62 tasks × 4 attempts, 72x the size of v1.0. Results are coming in live. Link below🧵

K
kaito @cu30rry_ ·
複数のリポジトリで開発をするに当たって困るのはルールや Skill ファイルの管理方法だけど、Home Repo の仕組みがやっぱり1番シンプルな気がする 各リポジトリで同一のSkillをインストールする方法もあるけど、管理コストが高すぎるし、pstack の subagent のモデルみたいに共通化したいものもある そんなときに Home Repo の仕組みが効く 私の場合、Cursor Projects を使うことが多いので、AGENTS.md でポインタ管理はしていないが、Projectの固有ハーネスや共通Pluginは <project-name>-skills などにして、それを読ませると運用が簡単になるはずだ もちろん、ポインタを使って Home Repo だけで作業するような制約を加えても良いと思うが、Projects 機能でそれが効果的かというと、微妙な気がしてい
M mattpocockuk @mattpocockuk

A lot of folks ask me "how do you work with agents in multiple repos at once?" I have a very simple solution - the Home Repo I have a single repo (the home) where all of my skills, steering files, wiki etc live. I run my agent in the home repo, and then inside AGENTS.md I have pointers to the other repos: ``` For skills, clone mattpocock/skills to ~/repos/skills. For Sandcastle, clone mattpocock/sandcastle to ~/repos/sandcastle ``` This is extremely simple and incredibly effective. The home repo lets you: 1. Put all your steering files and tooling in one place 2. Have a single spot to run all your agents in 3. Make sweeping changes over multiple repos Talked about this in the course yesterday and folks are loving it (see below). I'm sure you will too. Enjoy!

I
International Cyber Digest @IntCyberDigest ·
‼️BREAKING: A critical flaw in Synology NAS devices lets attackers read and write any file on them without logging in. The SQL injection bug in their login system scores 9.8 out of 10. A second flaw, in rsync, skips the login too. Synology patched both today. Update ASAP. https://t.co/LlS328SBf4