I know everyone is talking about Grok Bot, Instinct, and the latest model releases. But in reality, the highest leverage thing you can do to increase your team's AI productivity is invest in building a team-wide Compounding OS, ie. a company brain. In my latest video & essay, I offer a practical guide on how to do exactly that, covering: - Standardizing your team on an agentic platform - Building a robust shared skill library across your entire org - Making all your company context machine legible If you want to avoid the hype and focus on creating real AI leverage for your team, I'd encourage you to check it out. t.co/05MgTGxIBf

Instinct is cool for sure but I seem to be less wowed by it than many - maybe because Codex computer use is so damn powerful. I’ll also say, at least in the short term, agents that “have their own computer” are at a steep disadvantage relative to agents that can drive *your* computer, because your browser has your cookies and your unimpeachable residential IP. Right now remote control still feels disjointed in both ChatGPT and Claude, but that’ll get smoothed and streamlined, and when it does, texting with an agent that can drive your computer on caffeinate will be the ultimate personal assistant.

How could we massively boost productivity in health & care? Part two. A summary of comments on my last post about "Productivity starts with people", co-authored for @BMJLeader. These come from across multiple social platforms. Five themes stood out: 1. Who creates productivity? Richard Wylde argued productivity cannot be done to people. Those closest to the work see the opportunities and barriers most clearly, given trust, support and space. Ian Carruthers said the NHS needs to give people permission and authority to fix obvious problems. Marion Dixon asserted that good leadership which supports staff to “change their world” really matters. Dr Sadia Khan contrasted “control” that yields compliance, and “psychological safety and agency” that yield innovation. Laura Yearsley described people choosing the improvement areas themselves, with learning as the strategy in place of a detailed plan, with people from surgeons to nursing assistants choosing to make time for the work within demanding jobs. 2. Permission, time and support: David Groom noted change cannot sit on top of the day job. People want to be productive and need permission, structure, time and senior support. Jonathan Sunkersing observed progress comes from modest improvements done consistently, which makes it accessible to everyone. As a leader, Lesley Parkinson would agree the improvement gap with the team and hold those conversations weekly. Anthony Lawton recalled a manager complaining he produced ten ideas a week, and a finance director who asked how many were good. 3. The conditions for professional judgement: @farhadali said that protocols and prescribed ways of working protect safety; leaving no room for judgement kills the capability productivity depends on. Sarah Jenkins reported the instinct was to hand “productivity” work to finance colleagues, when culture and psychological safety affect us all. Vijaykumar V said without trust and relationship we can do nothing. Jennifer Carroll described creating the conditions for people to do their best work and connect authentically. 4. Shift from managing queues and capacity crises to designing for timely movement of patients, information and decisions: Chris Cooper argued for adopting economies of flow and features of the on-demand economy. Mike Chitty wants managerialism and care to become more comfortable partners. Richard Ogden would concentrate on the bottleneck. Martin Silcock asked who looks strategically at pathways. John Mortimer found the professional taking the call and visiting the person used 14% fewer resources with better outcomes. 5. Who and what we count: Sharon Anderson pointed to family caregivers as a second workforce, invisible in planning and productivity measures, creating hidden work on both sides. Tina Patel Gunaldo wants the same emphasis on what is going well as on what needs improving. Sonya Cullington said sustained AI use can erode the cognitive conditions improvement needs. Farai Pfende cautioned that extra effort and workarounds can make a struggling system look productive. Thanks to all who commented.

Here’s a site built on top of this viral post I asked @grok to scrape the replies, collect the sites and referrers and put that info into a directory I asked Instinct to build me a simple html site out of that content and deploy to my personal @Netlify and update the url All of it took 10 min & while I was doing dishes. Insanely wild times if you’re an AI-native builder t.co/DHC1bSaH1U

If you think instinct is good you’re gonna be fucking mind blown by what we’re building Private, open source and not trained on your data t.co/708TGF1Q6S

Instinct is a product case study to follow They're simultaneously 1. a disruptor - 10x better than any other personal agent in UX. In days it has captured >50% of my Agentic usage 2. ripe for disruption - on the user trust category. In a day, I halved my usage due to security

Who is Instinct AI (personal AI assistant) using for payments? Is this you guys at Stripe @jeff_weinstein? Surprisingly little/no info on the payment vault (who is processing, how PCI is saved, etc.)

question: is grok @bot and t.co/cf0AiTOZYc only exciting to people who haven't played with @openclaw or hermes?

My thoughts on Instinct and Grok @bot: both unexpectedly good, much better than ChatGPT Work I use Grok bot for work, and Instinct for personal as it’s one iMessage thread (tho it can parallelize itself) I’m amazed how good Instinct is at running long tasks and doing complex search/research My wishlist • be 10x faster • generate a UI to help showcase progress & results • use my phone‘s hardware, data and saved passwords/wallets • run locally to preserve privacy Now people with ADHD can get 10x more done

Sent this out to LP’s a few weeks ago. Manus and Poke are only the first to be recognized in this category. Over the past week, Instinct has made noise (for better and worse). My TLDR: The agentic harness will be the new wave of B2C and B2B2C in 2027. There will be genuine novel innovation here. Watch this space!

Update: this is the first time “AI” clicked for my dad Instinct drafted emails and filled arcane forms with data from previous emails and PDFs Sounds mundane to us but he signed up via text, did no setup, and it just works He’s not buying a new laptop because of Instinct. t.co/GvGDxiDtg7

I've been using Instinct as a daily agent for a few weeks and it keeps surprising me. It got me a prime-time table at a restaurant that was completely booked out by watching for reservation drops. Turned a month of my meeting notes into a clean monthly update. Built a pipeline in a single market that finds new deals drafts outreach for me to review while I sleep. The wildest one: it needed better lead data, hit the free tier limit on a platform, and asked if it could open a paid account with my card. I said yes. It signed up, paid, and kept going. That last one is where agents get interesting. Not just doing the tasks you assign. Hitting a wall, proposing a solution, and executing it with your approval.

I’m a convert. I’ve had friends all around the industry send me invite links to this thing. I was like “Nope!” I was extremely skeptical about it at first. But I’m in! It’s Part personal assistant, part therapist and always looking out for my back. Instinct has helped me so much in my day to day life. I’ve been a bit selfish in not wanting to share this but seriously everyone needs to be using this thing! I named mine Rocky Tars. If ya know, ya know!

You need to be on Instinct. And GrokBot. And Poke. What's your Townie look like? Are you using Hermes? Do you have a personal agent to manage your personal agents? No? I'll build it. What agent should I use to build it? Actually I'll just use an agent to decide what agent to use to build that agent.

Instinct/Pally feel so refreshing to what OpenClaw should have been. More than half the time with OpenClaw, I find myself debugging it to see why it’s crashed or long loops where the dashboard is just hung and unresponsive.

it shouldn't take a weekend to bootstrap a new agent/harness. grok bot, hermes, instinct, and many more to come. each one needs to get up to speed on you, your memories, skills, and automations. my system: keep all state outside the harness. I personally use @NotionHQ as a shared data store that both my agents and I can read/write to. state gets saved as work progresses, and bootstrapping a new agent is just connect notion mcp and go. can replace notion with anything multimodal that has a good MCP (github, obsidian, etc).

Spent time trying out both instinct and grok bot, this is the best articulation of the difference in how both feel. Both I think are a step change in how we interact with AI. t.co/QY6kLQs3hC

my friend's cat racked up ~$2k in vet bills this past year. he did everything right - filed the claims himself. insurance came back with more questions than he could answer. that was enough. he gave up. i sent him an Instinct invite. didn’t link his inbox (because of the discourse on this app). just gave it the invoices, records, and his pet insurance login. it matched every invoice to his policy and refiled the claims. here's the uncomfortable part: a huge chunk of that industry runs on exactly this kind of friction. Instinct, Grok Bot, and tools like them just deleted the admin tax they're priced on. we’re still underestimating what this does to every business model built on friction.

I bet there’s a scammer out there scraping Instinct power users from X and sending them phishing emails