Daniel Shapiro

The PA rn (shorts safu)

@shaundadevens @tradexyz sheesh

SpaceX projected to IPO at $2.5 trillion, set to overtake $TSM to become the sixth largest company in the world, per @tradexyz Pre-IPO Market estimates.

@doodlestein thanks for saving me from 600 jeff

Been trying to give the grok coding harness a chance, which is tough because I worry about it doing dumb stuff and screwing up my code. And it's a super high bar because Codex and Claude Code are already working so damned well. And even Gemini-CLI is good for code review. Do I really need another coding agent? Anyway, I decided to give it a shot anyway on a new project. So far, it's really not great. It's exhibiting behavior that I haven't seen from the frontier models/harnesses in a super long time, namely repeatedly saying that it's going to do all this stuff for me but then never actually doing much of anything. At this point it appears to have negative value to me because I feel the need to babysit it to prevent it from doing dumb stuff, and that's basically a deal killer when you have many dozens of agents to tend to at any given time. This all sort of goes back to what I've been saying about the brutality of the Pareto frontier. I joked recently about the line by Alec Baldwin in Glengarry Glen Ross: "As you all know, first prize is a Cadillac Eldorado. Anybody wanna see second prize? Second prize is a set of steak knives. Third prize is you're fired." I already have a third place agent, Gemini-CLI, and it only gets to do code review. Fourth place definitely seems to be "you're fired," for me at least.

@shaundadevens great stuff Shaunda!

@kaiynne @infinex 100% agree with all of this, nice post! Have you tried Factory Droid missions? Personally think it's the sota long running harness

@0xcarlosg great stuff @0xcarlosg

@Kunallegendd Great stuff Kunal! Love the breakdown of the 4 bot cohorts

@EffortCapital @kevinperron01 Nice.

@kevinperron01 Nice.

Nice.

@yacineMTB Goat mode

hero

@doodlestein @steipete how many tokens / what % of your 200/m plan budget has this run used? cheers!

@steipete Yeah, it's pretty nuts.

/goal + GPT 5.5 is amazing. I can now plan really extensive refactors with e2e tests and it just works.

@cognition @cerebras 😤

Intelligence at 1000 tokens per second, right in your terminal. Now available with SWE-1.6 Fast, powered by @cerebras. We're giving the first 100 people who respond a free month of Max to try it out.

@EffortCapital jm

WOW. Just use Blockworks IR. jito morning (jm) to all of those that celebrate.

@shaundadevens options are endgame. great report shaunda!

@thsottiaux @jxnlco a built in pdf parser

We are seeing strong traction and working to improve Codex for scientists across mathematics, physics, chemistry, biology and more. What do you wish it were capable of that it cannot do today?

B Cherney at the Sequoia talks said he has dozens of agents controlling 1000s of subagents I have experimented with the main agent being the orchestrator and offloading all edits to worker subagents but I agree with you it's just slower and not as good Curious if you have done any testing on this. Have an eval built out myself but not done with it yet

@KentonVarda You almost never want subagents in a world where autocompaction is as good as it is

My agents are always spawning subagents and telling them to do exactly the thing I told the main agent to do, except with a less-clear prompt. I feel like this isn't helping.

@EnoReyes love how the new plans work! would be great to see updated evals so I can bull post droid to everyone docs.factory.ai/benchmarks/ter…

More ways to Droid!

@0xSero cheers! if you like warp would recommend giving Zed a try. requires a little more setup out of the box getting the agent layout right, but you get a better editor than warp

@_dshap yes

My new favorite workflow =-=-=-=-=-= Pre-requisites 1. Setup Warp (either clone or use the app) 2. Setup Vibeproxy 3. Use Droid 4. GPT-5.5 =-=-=-=-=-= Workflow: - left bar for project management - files + git view - prooompt GPT-5.5 - Let Droid do your work

@blockworksres great dash

Big moment for @Blockworks Crypto is entering a new phase, and the winners will be the teams that help the market mature: better data, better transparency, and more trust between issuers and investors Proud to be part of this team and excited for what’s ahead

@realsigridjin @z Have you tried symphony yet? Played around with it last night and kind of felt weird. Still feel like you need an agent to create the issues for you in Linear otherwise they aren't detailed and specific enough. Early days

@z (can feel aura) says linear is new state machine for agents state machine is the spine of agents without state machine agents will lose the discipline everyone doing vibe coding, but you will be failed in production just with the vibe thats why we need ultrawork ulw