Devon Kelley

I had the pleasure of joining @Niklas_Sikorra on Venture Scale Podcast 🚀

I taught a 3 hour workshop on @openclaw @kalibr_ai and autonomous agents yesterday at @frontiertower 👾

@Jason I’m in

We started an AI founder twitter group... reply with "I'm in" if you're a founder and want to be added

Hey SF, come get set up on @openclaw on April 23rd! @chrisivester @colin_lowenburg @Jhillika and I are hosting at @frontiertower Sponsored by: @nebiusai @kalibr_ai and Dialog Claw-Curious? Get Your OpenClaw Agent Running in One Night. Join me! luma.com/Claw-Curious-W… via @LumaHQ

@hustlin_heev Dead 😂 what a bad bitch

LMAO my cleaning lady just walked in the bathroom while i was SHOWERING 😂 When i got out, changed, fired up google translate and showed her: "Sorprendida no estás muerta después de ver eso" whcih means "surprised youre not dead after seeing that" with a smile on my face and she took my phone and typed in "no my life is better" i guess i still got it jk jk

The `claude -p` subprocess approach still works. You just can't let any OpenClaw-branded strings reach the API. It's a cat-and-mouse game. They'll likely add new fingerprints. The fix is knowing how to find them: binary search the system prompt, test each half with `claude -p --append-system-prompt`.

the Claude Code workaround might work short term but you're still burning tokens on tasks that don't need a frontier model. I had the same problem and ended up just routing by task type. low reasoning stuff goes to cheap models, Sonnet only when it actually needs it. cut my burn 75% :)

Ouch, Claude tokens are expensive when paying for them as extra usage Trying to configure OpenClaw to use Claude Code instead of doing things directly through the API I believe that should effectively emulate using Claude a code directly and have it included in Pro subscription

@hustlin_heev Ngl I’m running like 8 different agents on OpenClaw and it’s amazing

funny seeing all these podcasters talking about Openclaw: • let it read ur calendar and send you a summary of your day • read hacker news send u top 5 updates to stay up to date on AI Bro idc lol Who is using Openclaw as their business analyst sending it hella unstructured data from disparate data sources to analyze ur funnel, make insights and content, and help u drive that sweet savory MRR Some good ones i found so far were Larry who automated tiktok posting and of course @jamesdevonport who made a marketing manager (seo, social media manager, etc.) We need to see more detailed videos in this space! Anyone have any recs?

"CLAUDE I ALREADY ASKED YOU 3 TIMES AND YOU KEEP GETTING IT WRONG, THINK LIKE A SOFTWARE ENGINEER AND FIX THIS BUG, MAKE NO MISTAKES"

Congrats to InspectGuard for winning the @kalibr_ai track at @FundingCommons hackathon!

@julianweisser @stefanoscalia @FedericoNoemie 🤣

@stefanoscalia @FedericoNoemie > I’m very interesting You’ve done more to prove the opposite by saying this. That’s not to say what you claim is necessarily false, but you’ve created an additional data point future proof will need to overcome.

I’ve yet to meet an interesting person who describes SF as boring.

this is the future of saas: one backend with two front doors. one for humans. one for their agents. same building. your users click buttons. their agents call your api. both do the same things. why: 1. agents are the new power users. people are already wiring claude code, codex, and openclaw into their workflows. if your product doesn't have an api, you're invisible to them. 2. no drift. one canonical api layer means one source of truth. ui and api never diverge. you must support agent to support human. 3. skills is the new distribution. when your api works with any agent, agents find their way in themselves. your product becomes a tool for their agent, not a tab a human have to visit. build for both from the start.

@Jbm_dev Passion for sure 🔥

on vacation right now woke up before sunrise walked 2 miles to a cafe in a town i don't know just to get a few hours of building in is this passion or a problem?

@nur_daulet_ It is gonna be super fun! :)

landed in SF! here for 2 weeks smth tells me its going to be super fun any fellow investors around?

@rishpoddar @elvissun @posthog Love the tool call visualization that’s cool! Are you not using adaptive routing as well? Your agents would avoid most of the issues you’re having to fix/optimize

@elvissun @posthog for LLM observability. github.com/rishabhpoddar/… for visualising agent tool calls as a graph (full disclosure that it’s created by me)

anyone have customer-facing agents running in production? what do you use for monitoring/logging/tracing?

@elvissun @elvissun I hope you’re not still digging through logs to make fixes your agents could have avoided using adaptive routing… are you? 😱

@sodofi_ @frontiertower There are other girls at frontier? 😂 let’s be friends!

my cowork setup when its sunny in sf

@Chiragagrwal7 Job market so bad the risk adjusted return of founding a startup actually looks sensible

job market so bad I'm actually pursuing my dreams

The first vertical festival for human flourishing in the age of AI. 🌸💻 This weekend, 1,000+ researchers, builders, founders, funders, and artists are taking over @frontiertower in SF for 36 hours of programming, experimentation, and collective imagination across AI governance, open-source infrastructure, and coordination systems. World-class speakers, hands-on workshops, an overnight hackathon, live public goods funding experiments, a community biolab, a 4,000 sq ft maker lab, immersive art installations, and live music into the night—all across 16 floors of one building. 🌐 Conference & hackathon are supported by @protocollabs @elevenlabs @NEARProtocol @FilFoundation @OctantApp @hypercerts @humntech @bittensor @solana @activeloop @vesslai @velda_io @NomadicML @unbrowse & more. In partnership with @torproject @internetarchive @foresightinst @AAI_Society @ethereumsf @UNICEFinnovate @metagov_project @founding Participate: luma.com/ftc-sf-2026 Hack: luma.com/ftchack-sf-2026