Marko Stankovic

A good way to reflect on #January6th is to listen to this Rational Security podcast recorded live in DC during the insurrection. I still can't fully process that this happened to our capital. podcasts.apple.com/us/podcast/the… @shaneharris @benjaminwittes @Susan_Hennessey @tcwittes

@bballbreakdown Do you comment on actual bball anymore? 90% of your posts are defending refs and not talking about the actual tactics and strategies of the game.

Well, the foot to foot contact by Strus is the obvious foul. As for the grab by Brunson's right hand, there's something there but it looks like it dissipates quickly enough to allow them to play thru it

@MicheleRivaCode @neogoose_btw Piece of information you’re missing is that in some places it takes 1-2 years to get a consular appointment

@neogoose_btw You're right and that's how I interpreted it, my attorney is just being super cautious though. Recommended I don't do that just yet

@iam_elias1 Did AI write this? This tool has been out for almost a year... most open source agent harnesses are model agnostic. Do your research.

Jack Dorsey just fired half his company because of AI. Then Block open-sourced the AI agent that replaced them. Block's goose achieved approximately 4,900 stars in its first 14 days after launch, the fourth highest two-week star velocity of any open-source AI project released in April 2026. Goose is Block's internal AI coding agent, open-sourced under Apache 2.0. It is the tool that Dorsey's engineers used to replace thousands of lines of human-written code with AI-generated equivalents. Then Block decided to give it away. Here is what makes Goose different from every other AI coding agent. It is not a wrapper around a single model. It is model-agnostic, you plug in Claude, GPT-5, Gemini, or any local model and it works the same way. It runs entirely on your machine. It has persistent memory across sessions. It supports MCP for custom tool integrations. And the design philosophy is the part that turns heads. Goose is built around the idea that an AI agent should be able to operate for hours or days on a single goal, not just answer one question and stop. It handles errors, retries failed steps, checks its own work, and escalates to the human only when it genuinely cannot proceed. Block fired 5,000 people. Then gave you the tool. Make of that what you will. github.com/block/goose

@unrootdesign I think their tagline is perfection. Highlights their main value prop and core use case in one line. All you literally need to do is read this to understand what they do.

What I'd push back on: The headline is too long. Built to make you extraordinarily productive, Cursor is the best way to code with AI. That's two headlines glued together. Pick one. The "extraordinarily productive" half is way stronger on its own

Cursor is worth $29B. Their entire homepage is just one trick. but they do it 4 times in a row. And it's the smartest thing any AI startup is doing in 2026. Here's what's going on 🧵

@GergelyOrosz All they needed to do was ask the AI agent what is best practices for deploying in production — redundancy is a no brainer.

Sucks for an AI agent to delete the prod DB - with no way to back it up - and risk the complete rental business. But the blame sits with the dev who decided to delegate decision making to the AI agent, and then not review actions, just YOLO it. Time for a blameful postmortem...

@anagatakaya @bcherny @claudeai @AnthropicAI Yeah -- that makes a lot of sense. Obv claude code didn't think of load testing his own auth servers.

@MarkoInLA @bcherny @claudeai @AnthropicAI I think they invalidated all the oauth tokens to shake of the openclaw instances/harnesses and then everyone was trying to re auth. Also the 15s timeout thing is terrible.

Unable to auth using /login into Claude Code today. The auth screen on claude.ai is a bit slow and then I get this error after pasting in the code "OAuth error: timeout of 15000ms exceeded". @bcherny @claudeai @AnthropicAI This has been my view all morning

@bcherny @claudeai @AnthropicAI Finally showing on their status... status.claude.com I wonder if related to their crackdown on auth usage in other harnesses?

@bcherny @claudeai @AnthropicAI Looks like auth servers are completely down now... before they were responding slow. Now they're coming back with internal server errors.

@jessicamalnik @bgurley This! Ive learned the hard way one too many times…

@bgurley The law of diminishing returns. It applies to everything but big ad spend before PMF or with high churn is a special kind of hell for B2B businesses.

Paid marketing is the crudest game you can play. It’s admitting you have no creativity. And actually restricts your creativity. Fire those that want to spend more. abovethecrowd.com/2012/09/04/the…

@j_v_n_85 @TheStudyofWar Has to be a WWI cartoon

NEW: A Russian victory is an Iranian victory. Those concerned with the growth of Iran’s military power, ambitions, and aggression in the Middle East must recognize the degree to which Iran’s fortunes rise and fall with Russia’s. (🧵1/9)

@rohanvarma Markdown viewer/editor. Easily manage multiple agents within a single view — think tmux panels. Better terminal, mines crashed a few times and I’m not sure how to recover without quitting codex. Most importantly — speed! Claude code just feels much faster.

If you have tried the Codex App, but don't use it as a daily driver yet: What needs to improve to make it the primary place you work with coding agents?

@mog_russEN This is Serbian brass music — they’re kissing each other 3 times. Don’t think it’s Turkish.

🚨Turkey A girl stands at the door of the house preventing her only sister from leaving with her groom during the wedding ceremony until he pays the “fee” — one of the old Turkish traditions

@HyperSalesman Thanks for the info -- still wondering what the best way to fill in demo/firmographic data for 100Ks of contacts we have in Salesforce to help with segmentation and freshnesss. Currently using Apollo but looking at other solutions to layer.

Marko is thinking about data the wrong way. - Firmographics - New businesses - Job changes - Promotions - Industry verticals Never look for these things in ZoomInfo, Apollo, Seamless, or any other database. The better approach: 1. Find tools that specialize in each of these 2. Build a master account list 3. Segment the list by location, size, & signals 4. Find 5-15 leads per account 5. Run async multi-channel outreach How I do this: - SalesNav + Exa for account building - DiscoLike for account expansion - Waterfall for email + phone data - Custom signals using Google API (funding rounds, etc.) TLDR SalesNav is your North Star for list building. Then Exa. And those two can identify most of the firmographics and signals you need. DiscoLike is a must for account expansion. Helps you find accounts you never knew exist. And run ASYNC outbound. - 30 days cold email - 5 days cold DMs - 10 days cold call

@HyperSalesman How about if the gap isn’t emails but demographic firmographic data? New company, title, industry, etc?

Everyone should just vibe code their own data waterfall using all of their favorite data providers. This can be done in 30 minutes. My recs on data: 1. QuickEnrich 2. FindyMail 3. Forager 4. Hunter 5. Wiza 6. Icypeas 7. LeadIQ This will cost < $1,000 for almost infinite data. Append every search result into a database that records: - Lead name - Company - Data source - Location You can also get their LinkedIn URL, company info, buying signals, etc. by combining Google’s search API + OpenAI API for formatting and additional research. Just ask Replit to do this. It’s easy. Everything with Apollo sucks. Their sequencer, dialer, and data are all mid.

@hadlinho @LDiken @petergyang @bcherny Me neither… still getting prompts as well.

@LDiken @petergyang @bcherny I am not seeing that behaviour. I am getting asked about everything in the world remotely and all of my live sessions in CC were with dangerously-skip-permissions

Need dangerously skip permissions for mobile remote control @bcherny 🙏

@rafiulm_ @dr_cintas I can second that... definitely feels a generation behind. Sometimes it makes tool calling mistakes that haiku gets right.

@dr_cintas Don’t trust benchmarks. It’s not even close to Opus.

MiniMax-M2.5 went fully open source and it’s basically Claude Opus performance but 95% cheaper.. it scores 80.2% on SWE-bench verified. It's 3x faster than the industry leaders. 4 useful use cases: 1.  Build interactive prototypes

@joekent16jan19 Yes, published by the @WSJ, notoriously pro democrat crooked media... 😅

Like clockwork: DNI Gabbard finds 2020 election fraud, days later the WSJ publishes a cryptic hit piece from anonymous sources. This is the Russiagate/impeachment hoax playbook, ment to divert attention from the DNI’s fight for the truth. No one is buying this crap in 2026.

@victormustar Some people can't comprehend that UX is important...

Funny how you can do the same thing with a local model without sending all your data to a remote cloud.

@hxiao same -- been getting this all morning

it's over again - claude code overloaded error.