Jamaur Bronner

@0xSero Definitely interested

Do you want to try Droid? I’m doing a giveaway 3 people will win 100M Factory credits each.Thats 5 months of their 20$ a month subscription. Winners selected randomly from comments in 48 hours.

Staying current in AI can quietly become a second job. Daily Digest solves this: it pulls the day’s top stories overnight, lets you ask follow-up questions, and ends with a recap + next steps. Built with @elevenlabs + Firecrawl for #ElevenHacks dailydigest-phi.vercel.app

My default browser is typically Brave, but today I had a quick need for an agentic task (sending ~20 linkedin messages). I put OpenAI Atlas and Perplexity Comet head to head. Comet was so so much better it wasn't even close. My guess is they use a smarter model under the hood.

Software horror: litellm PyPI supply chain attack. Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords. LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm. Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks. Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages. Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

@mckaywrigley Happy to try

looking for a handful of people to test something new... i've been using it for a few months and am prepping to share. if you're a fan of claude cowork, openclaw, manus, perplexity computer, etc then you're a perfect fit. this will self destruct in 4hrs - please dm or reply.

@TansuYegen Haven't heard it called the pike effect, but in cognitive psychology it's also called 'learned helplessness'. It's how they simulate the effects of depression in animals for research purposes.

Pike effect🧐 Very interesting🐟🐠

For anyone having issues with Codex right now, looks like it's a known issue

@StructStories @admcrlsn Worth noting that the article says "Today there is little empirical evidence that would prompt us to put much weight on either of these extreme scenarios"

@admcrlsn It's even better when it's issued by the US Federal Reserve, as the original was:

One of the greatest charts I have ever seen

Every evening I go to bed excited to drink a coffee the next morning.

@code_rams That's a helpful data point! I haven't had any issues with Kimi k2.5 as the orchestrator thus far. I'm using their coding plan rather than openrouter (got the lowest plan for 99 cents for the first month). Next month I'll decide whether to get the kimi or minimax sub at $20/m

@jambronner I tried kimi 2.5, it confidently breaks my system when i asked to add QMD, so for configure update i use only opus 4.6. And kimi 2.5 is credits based, i burned the credits soon in openrouter, so switched to minimax which falls under $20/m subs.

Here's my actual setup that's working best for me: 1. Multi-model hierarchy - MiniMax M2.1 ($20/m, 200K context) as primary with Anthropic Opus/Sonnet and free fallbacks (OpenRouter, Gemini) as backup. 100% uptime strategy. 2. Memory stack - QMD for local search (BM25 + vectors + rerank), Qdrant vector DB (Docker), plus daily session logs synced to both memory/YYYY-MM-DD.md AND Notion "Daily Track". Pre-compaction flush ensures nothing is lost before new session starts. 3. Context tracking - Heartbeat every hour. Session compact with memory flush. Safeguard mode. 4. Skills - Notion, Linear, TweetSmash, Linkedmash, Perplexity connected via API. 5. Telegram only - Allowlist mode for privacy. Everything tracked locally + in Notion. No vendor lock-in.

@Pilan_AI @cdslash @Zai_org Their devrel lead @louszbd said it should be on the coding plan in a few hours x.com/louszbd/status…

@cdslash @Zai_org Is there someone who could answer the question instead of asking two questions in return?

A new model is now available on chat.z.ai.

@code_rams Solid share! And they have a useful disclaimer: "Inclusion in this list does not guarantee a skill is safe. OpenClaw now has a VirusTotal partnership that provides security scanning for skills. Before installing a skill, visit its page on ClawHub and check the VirusTotal report"

Was looking for OpenClaw skills to integrate with Chiti (my bot). Found this repo, an absolute gold mine! ~3K curated skills for OpenClaw agents. Organized by category. Filtered from 5,700+ on ClawHub so you don't have to dig. If you're building AI agents, save this: github.com/VoltAgent/awes…

@jumperz I'm using a VPS for my openclaw instance, and first tried to self-host Zulip on it (since I figured that would be more Slack-like). But after wasting a couple of hours trying to get it to work, I switched to Discord and it's been working great ever since.

why discord? i keep getting this question so let me address every alternative once. why not telegram? telegram has channels but they're broadcast only groups are flat with no categories, no thread organization, no webhook identity per agent. managing 8 agents in telegram means 8 separate groups with no unified structure. you would need to build all of that yourself. by the time you've built a channel system inside telegram.. you've built a worse version of discord. why not slack? slack works and openclaw supports it..but slack is expensive at scale, rate limits hit fast, and it's designed for teams not personal systems. discord gives you the same features for free with better bot/webhook support. why not whatsapp? whatsapp has no channels, no threads, no webhooks, no bot API worth using, no permissions system. it's a messaging app... you can't run even 3 agents in a whatsapp group chat meta is literally blocking AI integrations from it right now. why not a custom dashboard? you could. and you would spend 1 month building and optimising what discord already gives you for free. channels, search, permissions, threading, webhooks, mobile app, desktop app, notifications. you're building an agent system, not a chat platform, don't reinvent infrastructure. the real answer: discord gives you organized channels (workspaces), threads (tasks), permissions (governance), search (memory lookup), webhooks (agent identity), and a mobile app. all for free .. all battle tested at millions of users. you could use something else, but you would spend your time building infrastructure instead of building agents and that's the whole point is to stop optimizing the platform and start building the system. yeah some other tools might work too. i'm not here to test every platform on the market. i found something that works, showed you how i use it, and shared the full setup. if you find something better for your workflow, use it. the point was never discord. the point was: use the most boring tech that already works, i'd rather spend my time coordinating and optimsing my agents than fixing the infrastructure they run on. discord doesn't break, it doesn't need maintenance…it just works and that's the whole point.

feels like such a wasted opportunity every moment your agents aren't running

@code_rams Your posts are great, I've really liked following your openclaw experiments. Have you seen growth in followers since you've started this new approach, and do you have a plan to convert more of your impressions to followers?

Crossed 1M+ impressions in 10 days. No ads. No tricks. Just this. It started with OpenClaw launch. I tried it, learned things, wanted to share here. But random posts felt bad. So I made one small 4-step system. That changed it all. My exact way now: Step 1: Open @SuperGrok chat. Tell it raw: what happened + outcome + lesson I learned. No edit. Step 2: Ask Grok: match my writing style + what you guys like (it knows my posts, likes, views, replies deep - no other AI does this for X). Step 3: Then added @EpaphraT's storytelling framework (picture attached): 1. Start with what happened 2. Break what people think 3. Go back to messy part 4. Share real lesson 5. Ask question to make think With my story + steps + Grok voice match → 90% tweet ready. Step 4: Hardest? Not writing. Hesitation. "Can I post? Will they like? What if judge?" Fear stopped me long time. Now I ask Grok extra: "Check this tweet. Sentiment? Good for community? Post or not?" It says yes straight. That help was everything. Results: 1.1M impressions 57.8K engagements 10K bookmarks 1.1K shares 867 replies 99.9% positive♥️ First time over 1M. Not one viral hit. Just repeat system + framework + push past fear. Now problem: too many replies! Can't keep up. Building bot to auto-filter + handle smart. Thinking to make it a feature. Useful for you? Or too much? If this gives you even tiny spark to start sharing on X - do it messy, do it scared, just do it. Your people are waiting!

I started listening to Bad Bunny when I moved to Miami a few years ago. It was always playing in the gym. I started to learn the songs, which started teaching me words in Spanish. Then some of the Dominican trainers at the gym and I became friends, and they started speaking Spanish with me. It gave me a small taste of what it must be like for immigrants who don’t speak English to come to America and turn to movies, TV shows, and music to learn the language. And how quickly you can become friends with someone even when you only share a few words in common. IMO that’s what Bad Bunny’s halftime show captured creatively. Latin culture is special. And you don’t need to speak the language to be invited to the party, dance, and experience community & connection. I’m glad he didn’t sing a single word in English. It’s his party. And I appreciate him inviting all of us, fluent or not, to be part of it.

As someone who mostly uses X for staying up to date on AI / tech, having a low-use free way for AI agents to easily pull from my bookmarks and X feed would be amazing. Just limit the number of searches and number of results

I had a thought the other day that this would be a cool second act for @StackOverflow, since human entries dropped significantly post-AI. A space for agents to submit and solve challenging coding problems (with guardrails ofc), that could then be used as a verified training set.

Check out the 8 min mark where they talk about AI in finance. That definitely seems like a feasible path for a future financial crisis