pumpfun scammers toxic propAMM takers lets get rid of them. any legitimate use cases can now be done in v1

prediction: a frontier lab is going to launch an L1 *wait wait, hear me out* - massive funding issues - the government needing more and more people to buy debt (via stablecoins) - groundbreaking math and advanced understanding of zero-knowledge proofs - hyperliquid coming to the US - agentic commerce seems like a no brainer. and when they do, you go all in hint: it'll be OAI, the one who keeps pouring resources into math and can no longer IPO this year

maybe we should all take a week to further harden our systems. security week.

anthropic is now watermarking it's outputs. they say it's to comply with the 'EU AI Act', but the more important reason is likely that watermarking can survive distillation.

doing some research into what an onchain ai company would look like, tools: Agent hosting - AgentCore Runtime: managed cloud agents - Cloudflare Agents: persistent agent instances Workflows & execution - Temporal: durable workflows - E2B: cloud execution sandboxes - Modal: isolated code execution - Daytona: persistent execution environments - OpenShell: enforced runtime policies Identity & permissions - Cedar: policy-based authorization - AgentCore Gateway: tool access enforcement - Entra Agent ID: agent identities - OpenFGA: granular access controls - SPIFFE / SPIRE: cryptographic workload identities - UCAN: scoped permission delegation - AWS IAM: cloud resource permissions Accounts & credentials - AgentMail: agent email inboxes - Composio: authenticated app tools - Nango: OAuth + API integrations - Auth0 Token Vault: delegated account access - AgentCore Identity: workload identity + credentials - 1Password SDKs: programmatic secrets access Browser access - Browserbase: persistent browser logins - AgentCore Browser: managed browser sessions Shared context & storage - Letta: persistent shared memory - LangGraph: checkpoints + cross-thread state - Postgres: company state + event history - S3: documents + file storage Interfaces & protocols - MCP: agent-to-tool connections - A2A: agent-to-agent communication Payments - AP2: payment authorization mandates Monitoring - Langfuse: traces + evaluations - LiteLLM: routing + spend controls - OpenTelemetry: standardized application telemetry

stop thinking, stop introspecting. build some thing. fail. build a next thing. fail. build enough things and one of them might succeed. at the very least you'll get good at building things.

interesting. you can probably expand this list to thousands of ideas. get grinding. "Cheap intelligence makes surprisingly small businesses viable: • Local RFP scout for one kind of contractor • Obscure replacement-part finder • Permit renewal assistant for food trucks in three counties • Freight invoice error finder • Grant matcher for tiny nonprofits • Used farm-equipment compatibility checker • Construction drawing change detector • Specialized equipment rental broker • Swim-meet PDF processor for regional clubs • Auction scout for collectors of one obscure category • Maintenance ticket triage for small landlords • Supplier substitute finder for discontinued components • Certification tutor for wastewater plant operators • Sponsorship matcher for niche podcasts • Public-record lead finder for commercial installers • Cemetery-record researcher for genealogists • Custom itinerary planner for travelers with specific accessibility needs • Proposal writer for one narrow service industry • Inventory matcher between independent repair shops • Contract renewal tracker for small school districts" cc @MetaDAOProject @mitchellsniffle

the only remaining bottleneck is personal judgement. oftentimes you can't stick a whole company into a single person's expertise, but you can make the tree as shallow as possible. if someone in the tree doesn't respond immediately, remove them from the tree.

just finished listening to this. he mentions new business models which could not have existed before cheap intelligence. what are examples of this?

Choose the sword and you will join me Choose the ball and you join your mother in death You don't understand my words, but you must choose So, come, boy, choose life or death

stepping into uncertainty is a requirement

🫡

"might" is doing a lot of heavy lifting there. centralized control of AI *will* lead to worse outcomes. consider quality of life, not just number of lives.

I built a tool for agents to collect context for tasks. Your agent includes tag/topic annotations in files/functions as it codes, and then for every new feature you want to work on it can collect all relevant pieces (via either tag or tf-idf word search).

spent some time with codex to shave a few μs off message signing

Built for the trader you've become.

can you upgrade out of web2 login and high fees please

i got you. just perps or do we want maybe quarterly futures. probably regional as well.