William Estoque

1.2K posts

William Estoque

@westoque

https://t.co/xmAjVyjPPx creator. Partner @narralabs. CTO @wizestapp. CTO agencymvp. VP of Eng @guidevine (acq blucora). @outright (acq godaddy).

nyc Katılım Ağustos 2025

239 Takip Edilen208 Takipçiler

Sabitlenmiş Tweet

William Estoque@westoque·9 Şub

just released clawinit.com managed hosting for openclaw deployments. end goal is to have employee agents per deployment that are pre-configured. targeting entreprise which means getting security buttoned up with SoC2. this is not just a one-click openclaw setup, it's a click for as many employees as needed. bigger goal.

English

1.8K

William Estoque@westoque·6h

to make them 100% you have to solve for most of the unreliable cases. an example is having interactive inputs in the terminal, a model gets stuck now what? another is in the browser, it's waiting for an interaction to load and stuck. i'm doing this day to day. the way to know and solve for this is to add monitoring tools and human-in-loop to solve. the termina? you should be able to step in and solve and debug. for browser, have a browser to go into and control. all these unreliable cases i have already built regression on and i have a list of these things that make my harness much more reliable. my goal is to have RELIABLE AI employees that work 24/7.

English

Henry Mascot@iAmHenryMascot·6h

Cool I can use agents Cool they work Iike hell Next, How do they become 100% autonomous ? I have 150 crons, they are pretty wired up to do many things, still pretty dumb though. Next unlock …. Reasonable autonomy So I can remove myself totally Send help if you know

English

126

William Estoque@westoque·1d

@yenkel could potentially be. they have their own email so they could be invited easily. in the future they can be using their own company emails like agent-one@entrepriseco.com

English

yenkel@yenkel·1d

@westoque interesting so it’d be an enterprise jira account the the agent should access?

English

yenkel@yenkel·1d

wish time 🪄 name 2 dev products you wished your coding agent could easily create accounts for

English

1.5K

William Estoque@westoque·1d

@yenkel agentic development. just let them loose in jira tasks and they submit PRs in github as their own user not access tokens.

English

yenkel@yenkel·1d

@westoque interesting. what for?

English

William Estoque@westoque·1d

@yaghsizian @Shpigford @polsia i use AI everyday with these AI employee use cases and this is certainly true. it's not as perfect as it sounds. i'm not being negative and want it to work but it's just what it is atm.

English

Yags@yaghsizian·1d

@Shpigford @polsia Polsia is an absolute slop factory, the ARR claims are complete bullshit. I’d be surprised if users have <90% churn

English

228

Josh Pigford@Shpigford·1d

in today's random email from @polsia... the duplicate unsubscribe links...fine. i get it. templating errors happen, especially with email. but that magic link very much logged me into somebody else's account. 🫣

English

4.2K

William Estoque@westoque·1d

i'm calling it. we need agentgit the repository for agents that have agent policies in place to prevent slop. anybody building this? will put up some $ to help.

clem 🤗@ClementDelangue

Our biggest open-source repos are getting overwhelmed by AI slop which literally makes Github unusable (~a new pull request every 3 minutes). Fun new challenges in an agentic world!

English

William Estoque@westoque·1d

@oliverhenry @NickScheifler got banned with the 401 as well. only works in CC

English

Oliver Henry@oliverhenry·1d

@NickScheifler Is this true?

English

713

Oliver Henry@oliverhenry·1d

Has anthropic finally banned openclaw? Just upgraded my max plan to 20x from 5x. Went from working fine, to: run error: HTTP 401 authentication_error: Invalid bearer token Thought I'd treat Larry to more tokens, now i think i have to cancel my anthropic subscription and switch.

English

12.2K

William Estoque@westoque·1d

@hthieblot infrastructure for autonomous workforce. get an AI employee at a click of a button that have their own identity (email) and economy (cards, wallet).

English

Hubert Thieblot@hthieblot·2d

Looking for obsessed builders. I invest up to $250K first checks in: • Robotics, drones, space • Applied AI/ML, models • Dev tools and infra • Manufacturing & logistics, and more... DMs open or just reply here what you are building. Early > polished.

English

456

1.4K

100.3K

William Estoque@westoque·1d

@paulg @claudecademic i think it just depends on market demand. the neo-vintage market had good value prior. watches such as 3970, 3940, 5066, 5065 then were a steal.

English

128

Paul Graham@paulg·1d

@claudecademic Don't buy a Patek made after 1990. Then you won't have to deal with such tactics, and you'll end up with better watches too.

English

5.6K

Paul Graham@paulg·1d

I went to visit Subdial in London. Cool place with some good watches. I bought something so strange that till I saw it I'd never have believed it existed: an Ellipse pocket watch. I kid you not.

English

810

111.3K

William Estoque@westoque·2d

@UbermenschMind i coin this term vacadoro

English

130

Alexander 𖤓 Nietzschean Vitalist@UbermenschMind·2d

To avoid burnout use the 30/30 rule: After 30 minutes of work quit and disappear to Southern Europe for 30 years

Alexander 𖤓 Nietzschean Vitalist tweet media

English

2.1K

26.8K

William Estoque@westoque·2d

another thing you'll encounter once you have actual 24/7 working agents is they misbehave. notice this behavior where the agent is saying it selected sms but really it's still in device. thankfully we have implemented live terminal agent access so i can intervene.

English

William Estoque@westoque·2d

@adisingh wanted someone to build agentgit as a git repository for agents. anybody? better if has humans + agents working together and easier account creation.

English

248

Adi Singh@adisingh·2d

If you are looking for a startup idea and want to be rich, you should probably build in this category.

Ivan Burazin@ivanburazin

New category emerging: Headless SaaS Not infrastructure as a service / platform as a service Traditional software (Photoshop, Slack, Jira) rebuilt with agent-first APIs. - No UI - Programmatic access - Essentially the same product with different interface Entirely new business model.

English

124

37.3K

William Estoque@westoque·2d

@yongfook i mean probably better to paste logs into claude and say create tickets and potential solution for each and review each one by one. safer imo.

English

465

Jon Yongfook@yongfook·2d

If you haven't already, tail your server logs and filter by requests that take longer than 500ms. Copy paste it all into Claude and say "pls fix". Instantly faster app.

English

646

78.7K

William Estoque@westoque·2d

@rot13maxi too deterministic imo. the agent needs to be more free to do whatever is needed to the output.

English

Rijndael@rot13maxi·2d

building the dominos pizza tracker for agentic software engineering

Rijndael@rot13maxi

stripped back the UI while I work on the harness itself. Then will be adding shiny UI back. Interaction model first.

English

2.1K

William Estoque@westoque·3d

at launchfest. reach out if you're here

English

William Estoque@westoque·4d

@Shpigford @openclaw most consistent for us is to use the openclaw sandbox browser. is that what you're using? or the chrome extension?

English

132

Josh Pigford@Shpigford·4d

what's the most bullet-proof setup for getting @openclaw to consistently navigate/interact with sites/apps in the browser? it's exceedingly inconsistent for me. just feels like i must have it set up incorrectly. FWIW this is running on a Mac mini. Not using headless browser.

English

5.5K

William Estoque@westoque·4d

this guy agents!

Zhu Liang@paradite_

you want to build an agent? it’s easy, just learn: models, tools, prompts, harness, providers, api formats, multimodal, streaming, reasoning, sessions, compaction, handoff, token accounting, cli, mcp, embedding, skills, memory, abort controller, sse, process isolation, sandboxing.

English

William Estoque@westoque·5d

@snowmaker @jack getting stuck happens a lot more in UI land vs predictable backend code. you want an example? just try to ask it to copy a certain theme/design from one place to another.

English

163

Jared Friedman@snowmaker·5d

I realized something else AI has changed about coding: you don't get stuck anymore. Programming used to be punctuated by episodes of extreme frustration, when a tricky bug ground things to a halt. That doesn't happen anymore.

English

592

444

7.4K

904.4K

William Estoque@westoque·5d

bingo! this is where the battlefield is in agent-land. stability and reliabilty among other things. that's we're trying to do.

Henry Mascot@iAmHenryMascot

Most AI products are still just demos with better branding. The model is not the moat. The moat is the layer around it: memory, reliability, evals, fallback paths, distribution. If one hallucination breaks the experience, you didn’t build a product. You built a prompt.

English

Keşfet

@yenkel @yaghsizian @Shpigford @polsia @oliverhenry @NickScheifler @hthieblot @paulg