Mark. (@markmdev) header
Mark. (@markmdev) profile picture

Member of Technical Staff @a____t____g 9x hackathon winner SF

San Francisco Bay Area
Joined April 2025
Mark. retweeted
what’s one HARD thing you want to do this year? we asked our friends, then built misogi to help them do it ben is doing a 100-mile race. alvin will hit 100k mrr. josh is reading the odyssey in the original greek no new app. misogi turns your existing @Muse into a trainer, tutor, mentor that already knows you best what’s your misogi? reply and i’ll send you a plan to get started 🏔️
16
6
2
43
1,700
we built “let’s bump phones to exchange contacts” but for agents! now your @instinct / @bot / @muse (whatever you use!) can make plans, negotiate deals, and get what it needs from your friends/coworkers’ assistants it's “have your ppl (agents) talk to my ppl (agents)!” for everyone 🤝 built w/ @uncanny_guzus @ @ycombinator #MakeSomethingAgents hackathon (s/o @nemild) on top of @stripe (agentic p2p payments) + @manufact (mcp relay server) 🙌 Who will you bump first?
46
16
4
256
19,674
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
20,321
166,501
39,479
798,843
175,847,160
Mark. retweeted
This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.
9,132
35,051
26,632
337,615
139,553,410
i'll never forget that time someone demo’ed me a language model at a hackathon in 2022 and I thought to myself "this is just GPT-3 but worse" and it turned out to be Claude 1
60
96
6
11,916
387,316
Mark. retweeted
WE RESEARCHED ALL 6,077 YC STARTUPS We wanted to know: do these startups share anything in common? And how have trends shifted over time? Here's what we found:
1
2
5
252
WHERE ARE YC FOUNDERS NOW? @_javokhir and I were curious where they go "after the startup". So we ran Deep Research on ALL 11,398 founders who have ever gone through YC. Here's what we learned: joinedanthropic dot com
Personal update: I'm taking a leave of absence from YC to join Anthropic. I'll be working with @NotTomBrown on the compute team. Powerful AI has the potential to improve the life of every human on earth and, as we enter the early stages of recursive self-improvement, availability of compute becomes one of the most important issues to solve. I'm excited to get started 🚀
9
7
8
134
67,740
Mark. retweeted
I embedded all ~6,000 launched Y Combinator companies by what they actually do, and mapped idea-space.
3
3
6
340
A great prompt from @BusMark_w_Nika: What tech or product was too early for its time and worth rebuilding today? Here are a few that came to mind: 1/ "tools for thought" (huge in early 2020s), seems like llms were the missing piece? instead of manually curating, I just want my ai to manage my personal knowledge. rly liked @MuseAppHQ by @_adamwiggins_ 2/ magic leap -- anyone remember the whale demo? 3/ pebble -- excited for pebble time 2 by @ericmigi! 4/ AI shopping -- fetchr (w23) tried it (@calvinchen @calixo888), rly like the fresh take @bindra_dhruv is trying w/ @zamana_hq, think they're finally gonna nail it! 5/ @Detour by @andrewmason -- imagine walking through a city with a real-time voice agent guide that knows where you are, sees what you see 6/ lots of OLD ideas will be revisited in chip design post-Dennard scaling: photonics, neuromorphic / bio-inspired, approximate computing, non-Von Neumann, etc. 7/ ok this one might be a bit unhinged, but @theranos by @ElizabethHolmes was a great idea? why am I still doing blood tests twice a year instead of weekly at home? Which startups would you revive?
1
1
6
338
Mark. retweeted
Introducing Straude: Strava for Claude Code 🏃‍♂️ One command logs your token output, spend, streaks. Sync to your straude dot com profile and compare your pace to the top AI builders worldwide. Available now: ~npx straude Check it out & upvote us on Product Hunt (#10) ↓
5
3
4
20
7,597
Mark. retweeted
Someone built a leaderboard for tokenmaxxers and the top user burned $10k on Claude Code this month. One CLI command syncs your token spend, models, and streaks. See what your friends are pushing. Where do you rank? npx straude
4
6
31
2,444
Introducing The Prometheus List: top companies with unlimited token budgets you should work at 🔥 It’s 2026, an unlimited token budget matters more than unlimited PTO. If your manager is asking you to "chill out with the agents" or "just use Sonnet," it’s time to look for a new gig. feat. @tryramp @NotionHQ @vercel @resend @Railway @browserbase @genspark_ai @ScoutAI_ @ExtendHQ & more straude dot com/token-rich
5
2
1
16
1,522
And an even better day to bunx straude 🤗
A beautiful day to Claude Code
1
6
468
Mark. retweeted
feels like such a wasted opportunity every moment your agents aren't running
489
270
139
4,430
531,810
Mark. retweeted
Introducing Claude Opus 4.6. Our smartest model got an upgrade. Opus 4.6 plans more carefully, sustains agentic tasks for longer, operates reliably in massive codebases, and catches its own mistakes. It’s also our first Opus-class model with 1M token context in beta.
1,656
4,613
2,897
38,580
10,630,912
Mark. retweeted
We're experimenting with ways to keep AI agents in sync with the exact framework versions in your projects. Skills, 𝙲𝙻𝙰𝚄𝙳𝙴.𝚖𝚍, and more. But one approach scored 100% on our Next.js evals: vercel.com/blog/agents-md-ou…
77
135
97
1,546
471,133
Progressive disclosure is not reliable because LLMs are inherently lazy. "In 56% of eval cases, the skill was never invoked. The agent had access to the documentation but didn't use it." Vercel ran evals on Next.js 16 APIs that aren't in model training data to test whether agents could learn framework-specific knowledge through Skills vs. persistent context. Skills are the "correct" abstraction: package domain knowledge, let the agent invoke it when needed, minimal context. The agent decides when to retrieve. They work well WHEN the user triggers them; otherwise, LLMs just ignore them. Vercel's benchmarking is the first experiment of this kind I've seen, and it's actually interesting. - Baseline (no docs): 53% - Skill (default): 53% - Skill with explicit instructions: 79% - AGENTS[.]md with 8KB compressed docs index: 100% The skill approach assumes agents reliably recognize when they need external knowledge and act on it. They don't. "You MUST invoke the skill" made agents read docs first and miss project context. "Explore project first, then invoke" performed better. Same skill, different outcomes based on prompting. The winning approach removed the decision entirely. An 8KB compressed index embedded in AGENTS[.]md, with one instruction: "Prefer retrieval-led reasoning over pre-training-led reasoning." Two agent design learnings: 1. Passive context beats active retrieval for foundational knowledge. Don't make the agent decide to look things up, make the index always present. 2. Compress aggressively. Vercel went from 40KB to 8KB (80% reduction) with zero performance loss. The agent needs to know where to find docs, not have full content in context. The gap between "agent can access X" and "agent will access X" is larger than we assume. I keep seeing similar findings across agent architectures. Kimi Swarm's orchestrator is trained specifically to avoid sequential execution. Without training, orchestrators default to serial processing, planning a list of steps and executing them one by one. It's the EASY path. The agent defaults to the lazy path: hallucinating from training data rather than retrieving docs. Passive context removes the choice entirely; the agent doesn't decide whether to look things up; the index is already there. We keep finding that the "smarter", more autonomous design (let the agent decide when to X) underperforms the "dumber" design (always X, or structurally enforce X).
We're experimenting with ways to keep AI agents in sync with the exact framework versions in your projects. Skills, 𝙲𝙻𝙰𝚄𝙳𝙴.𝚖𝚍, and more. But one approach scored 100% on our Next.js evals: vercel.com/blog/agents-md-ou…
63
90
26
1,110
198,468
Mark. retweeted
WOW!!! If you have semantic memory tied to your UserPromptSubmit hooks, you MUST ALSO include it in your PreToolUse hook. I promise you - it will be an absolute GAME CHANGER. It will put your efficiency levels are over 9,000 (*vegeta voice*). How many times have you sat there, watching Claude code go through an extended workflow, just to notice it start to go down a path you just KNOW will be error filled - and subsequently take it forever to FINALLY figure it out? The problem with relying strictly on the UserPromptSubmit hook for semantic memory injection is the workflow drift from your original prompt. The memories it injects at the initiation of your prompt will be less and less relevant to the workflow the longer the workflow is. Claude has a beautiful thing called thinking blocks. These blocks are ripe for the picking - filled with meaning & intent - which is perfect for cosign similarly recall. Claude thinks to itself, "hmm, okay I'm going to do this because of this", then starts to engage the tool of its choice, and BOOM: PreToolUse hook fires, takes the last 1,500 characters from the most recent thinking block from the active transcript, embeds it, pulls relevant memories from your vector database, and injects them to claude right before it starts using its tool (hooks are synchronous). This all happens in less than 500 milliseconds. The result? A self correcting Claude workflow. Based on my testing thus far, this is one of the most consequential additions to my context management system I've implemented yet. Photos: ASCII chart showing the workflow of the hook, and then two real use-cases of the mid-stream memory embedding actually being useful. If you already have semantic memory setup, just paste this tweet and photos into Claude code and tell it to implement it for you. Then enjoy the massive increase of workflow efficiency :)
26
50
7
664
60,201
Your timeline just exploded with indie Roblox devs making millions. But did you know there are creators earning just as much making cute lil stickers for an Asian messaging app (LINE)? @fahmitech wanted our own stickers, but we can't draw, so instead we made AI Stickers: Upload a selfie, get custom sticker packs ready for the LINE marketplace. Vibecoded this morning using @cerebras @cline GLM4.7. Check it out!
20
21
2
58
1,247