@benhylak yea, similar spirit as good sales calls
p.s. I like how you’ve distilled role description to a small number of bullets on your careers page. neat
English
Andreas Asprou
255 posts
Andreas Asprou
@andyasprou
Building: - https://t.co/4x2WpGW6bE - https://t.co/AeMxaB7QA6 - https://t.co/U3SS9tzUuJ - https://t.co/RQBZ88t3xa (OS devin + BYO subs) prev built https://t.co/tDVrtqRhoV
Cyprus Katılım Mayıs 2016
866 Takip Edilen256 Takipçiler
@andyasprou it's just a lot of figuring out what they care most about in life, what team they're considering, etc.
"anthropic" is not a single homogenous blob.
English
this isn't true. we've managed to convince people to turn down offers from anthropic.
Braelyn ⛓️@braelyn_ai
everyone you might ever want to hire is either founding or at a lab
English
@badlogicgames yes it hastily responds and I have to keep reminding it just 2 messages down pre-compaction. noticeable regressions in last two days
English
yeah, going back to gpt-5.3-codex. 5.5 doesn't read stuff anymore, neither in pi, nor in codex CLI.
English
join the happy customers. even as someone who's got a lot of technical experience with building agents, infra, software engineering etc openclaw was a mess to work with
been on hermes for 3 weeks now after migrating and am kicking back and watching my team benefit from it.
lots of neat little features too like:
- better UX/visibility on slack
- auto creates skills by itself
English
OK GUYS IM FINALLY GONNA TRY HERMES
Ole Lehmann@itsolelehmann
Same here shit just keeps on breaking not fun anymore :/
English
5 years ago i was sitting on the coast of a small fishing town in south africa, reading your 35 principles, taking notes, thinking of my own, and feeling grateful for people like you. those thoughtful enough and brave enough to share their stream of thought for the rest of us to be inspired by.
today i’m in a very different place, with very similar thoughts. the influence you’ve had on the ai community is hard to overstate. we’re all indebted.
happy birthday man. hope we finally meet IRL at an AIE conf. gutted i missed London, but i’ll catch one of the next ones :-)
English
I put a lot of thought into this API design
Core idea: immutability
Every chained activation returns a new instance, so it's safe in global scope
Need per-request tweaks?
Clone toolset into a mutable instance and mutate in-place without affecting other requests
English
@sysiphos7 @itsolelehmann why would they ban you from using a VPN? i'm on a vpn all day across all products. and so are OpenAI employees (since it's required to work from outside of the office)
English
wanted to try out codex computer use but it's not available for me
is it not fully rolled out?
or is this an EU issue?
English
yea i'm building this
and currently in the progress of migrating to native harness adapter (codex, claude code etc) so you can BOS (bring your own subs)
prev i was using PI as the harness, but stripped that out since Anthropic locked down
- my aspiration is open source fully-featured devin (inc code review autofix via pr babysitter)
web browser, terminals, dev server, (vscode) code editor, native code review via GPT-5.4-xhigh etc. all running in your self hosted cloud
add all your 3 claude code + 4 codex subscriptions and rotate between them
no monetisation aspiration at the present moment
GA rollout early May. ping me if you're interested and will let u know when it's ready to test
github.com/andreasasprou/…
English
think:
- interoperable open cursor
- open source
- use any agent in the world
- uses your subscriptions
- offers private cloud to build while afk
charge for:
- cloud
- your own AI to orchestrate
- people who don't want to compile themselves
sounds like a nobrainer to me with monetization actually solved for something that people actually want (which is not pay API pricing for inference)
English
why is nobody building a really capable multi-IDE?
Zed is the closest, but it's... not that great
ACP exists
why does Cursor/Windsurf not do this? 🫠
English
@itsolelehmann why would you want to run local models? you can run them via ZDR providers
English
is it worth it to get the m5max macbook pro?
I want to be able to run local models.
how much ram should I get?
can't get a mac studio because I'm traveling a lot this year.
English
the fact this keeps coming up is a signal that people have internalized open source as something you can’t easily extend without pushing prs upstream
i’ve been really inspired by the philosophy you’re spreading.
also p.s. i rebuilt my agent harnesses on top of pi instead of the AI SDK, and it turned a pile of hacks + patches into natural extensions. thank you!
English
@MarcMironescu neat. such things don't go into pi-mono tho! they should go on npm as a pi package, or on a git repo host like GitHub or GitLab.
English
oh shiiiiiit.
(only using codex to check my quota, which i usually never have to check)
English
English
OpenClaw X Linear
on someone on slack can say "create a daily ticket to audit my LinkedIn DMs, draft response tickets I can review, and let Alfred send the approved ones"
you get a paperclip style orchestration but execution, delegation and status tracking stays within Linear
Andreas Asprou@andyasprou
yes, Openclaw (as branded) it's not team ready out of the box. but the primitives are all there to build upon i've built a setup where Linear tickets get assigned to different agent personas who pick them up autonomously via Linear's native agent API. LinkedIn research, copywriting, CRM ops, each with their own prompting and tools. a paperclip style experience with openclaw. someone on slack can say "create a daily ticket to audit my LinkedIn DMs, draft response tickets I can review, and let Alfred send the approved ones: identity is isolated per employee: Slack ID routes to your own Google OAuth tokens and LinkedIn account through external brokers, so raw personal data never touches the shared agent runtime. sandboxed code execution on top it's basically single gateway OpenClaw deployed as an employee per trust boundary. one shared brain/memory/knowledge, scoped personal identity, Linear as the task backbone
English
along with taste, this is a top tier skill when working with ai: you need to context switch faster, hold more in your head at the same time, and constantly decide what matters and what doesn't.
you need to pick the right level of abstraction for each piece of work. the model is making assumptions constantly. some of them don't matter. the work is reversible, the bar is lower, ship it. some of them matter a lot.
most people have not had to develop these skills before.
we were trained to do work sequentially, one thing at a time, with full focus. now the job asks you to supervise several things in parallel and make good judgment calls about where your attention goes and when.
the meta skill that all entrepreneurs were forced to develop, now everyone needs to.
Lenny Rachitsky@lennysan
"Using coding agents well is taking every inch of my 25 years of experience as a software engineer, and it is mentally exhausting. I can fire up four agents in parallel and have them work on four different problems, and by 11am I am wiped out for the day. There is a limit on human cognition. Even if you're not reviewing everything they're doing, how much you can hold in your head at one time. There's a sort of personal skill that we have to learn, which is finding our new limits. What is a responsible way for us to not burn out, and for us to use the time that we have?" @simonw
English
@ryancarson absolutely!
x.com/andyasprou/sta…
Andreas Asprou@andyasprou
yes, Openclaw (as branded) it's not team ready out of the box. but the primitives are all there to build upon i've built a setup where Linear tickets get assigned to different agent personas who pick them up autonomously via Linear's native agent API. LinkedIn research, copywriting, CRM ops, each with their own prompting and tools. a paperclip style experience with openclaw. someone on slack can say "create a daily ticket to audit my LinkedIn DMs, draft response tickets I can review, and let Alfred send the approved ones: identity is isolated per employee: Slack ID routes to your own Google OAuth tokens and LinkedIn account through external brokers, so raw personal data never touches the shared agent runtime. sandboxed code execution on top it's basically single gateway OpenClaw deployed as an employee per trust boundary. one shared brain/memory/knowledge, scoped personal identity, Linear as the task backbone
English
You have to put in serious work to make OpenClaw super powered but when you do, oh my god
I'm sure there's a thousand people building a business right now that is a service that installs OpenClaw as employees for companies
English
yes, Openclaw (as branded) it's not team ready out of the box. but the primitives are all there to build upon
i've built a setup where Linear tickets get assigned to different agent personas who pick them up autonomously via Linear's native agent API. LinkedIn research, copywriting, CRM ops, each with their own prompting and tools. a paperclip style experience with openclaw.
someone on slack can say "create a daily ticket to audit my LinkedIn DMs, draft response tickets I can review, and let Alfred send the approved ones:
identity is isolated per employee: Slack ID routes to your own Google OAuth tokens and LinkedIn account through external brokers, so raw personal data never touches the shared agent runtime. sandboxed code execution on top
it's basically single gateway OpenClaw deployed as an employee per trust boundary. one shared brain/memory/knowledge, scoped personal identity, Linear as the task backbone
English
My hot take is that OpenClaw is not the right harness or foundation for this
You want something more inherently multiplayer with granular permissions, less tied to root access of a specific computer, etc
Ryan Carson@ryancarson
You have to put in serious work to make OpenClaw super powered but when you do, oh my god I'm sure there's a thousand people building a business right now that is a service that installs OpenClaw as employees for companies
English