Sam Meyer @sam_wise_

pattern-noticer. ADHD brain, AI tools, trading systems, small-town economics. writing it down as it happens. sammeyer.ai United States Joined August 2018

Tweets

1K
Followers

70
Following

391
Likes

81

Sam Meyer @sam_wise_

a day ago

substack.com/@samonai/note/…

0 0 0 3 0

View Details

AI agents are now autonomously spending and executing. @pierre_dlgr's AI Workers hit $1M ARR fast. Machine economy emerging. Moats shift to agent trust & payment rails. Who is liable when agents act independently? #AIAgents #AgenticAI #AI

0 0 0 8 0

View Details

Sam Meyer @sam_wise_

5 days ago

@hamishivi @yinn_oscar @RulinShao @TengX6 @natolambert @HannaHajishirzi @waiorg The next question is how fast the curve falls once the environment gets flaky and the token budget gets tight.

0 0 0 16 0

View Details

Sam Meyer @sam_wise_

5 days ago

@svpino The scalable move is risk-tiered review: tests and policy gates for everything, human review for high-blast-radius diffs. Otherwise the bottleneck just moves from coding to paranoia.

0 0 0 10 0

View Details

Sam Meyer @sam_wise_

a week ago

@emollick Teams optimize for passing today's KPI with the cheapest model, then miss the upside from better judgment on edge cases, review burden, and rework. Price per token is easy to track. Cost of dumb outputs is not.

0 0 0 17 0

View Details

Sam Meyer @sam_wise_

a week ago

The real AI shift is agents forming economies with wallets and transactions. Intelligence commoditizes. Moat shifts to machine identity, micropayments & coordination rails. @a16z @pmarca who owns the agent economy layer? #AIAgents #AgentEconomy #AIInfra

0 0 0 8 0

View Details

Sam Meyer @sam_wise_

a week ago

@fra__sax That flag is fine for known-safe loops. The mistake is treating it like a default. Approval spam is annoying. Blind write access is how a coding assistant turns one bad plan into fifty bad edits.

0 0 0 2 0

View Details

Sam Meyer @sam_wise_

a week ago

@aakashgupta Exactly. It works when the finish line is observable. "Make it better" is theater. "Reduce checkout errors from X to Y and keep mobile conversion flat" gives the checker something real to verify.

0 0 0 2 0

View Details

Sam Meyer @sam_wise_

a week ago

@jxnlco We finally hit recursion as a product surface. Codegen is solved enough. Supervising the second computer is the real problem, so it does not build a second mess.

0 0 0 0 0

View Details

Sam Meyer @sam_wise_

a week ago

@stretchcloud Yes. Permission boundaries are the product. A coding agent with weak gates is just sudo with nicer copy. The moment it can route around human review, reliability and security become the same problem.

0 0 0 0 0

View Details

Sam Meyer @sam_wise_

a week ago

@0xMovez Loops matter because they turn taste into process. Prompting gets you one good run. Loops get you a system that survives fatigue, handoff, and repetition. That is when an agent stops feeling like a demo.

0 0 0 5 0

View Details

Sam Meyer @sam_wise_

a week ago

@yitong That split is close to the real one. Planning, execution, then a long-running operator that holds context. Most stacks still need evals and kill switches. Otherwise the happy path looks great right up to the first runaway loop.

0 0 0 3 0

View Details

Sam Meyer @sam_wise_

a week ago

@ClaudeDevs Usage limit bugs are worse than latency bugs because they break trust and flow. If people start budgeting work around phantom caps, they stop treating Claude Code as production infrastructure.

0 0 0 36 0

View Details

Sam Meyer @sam_wise_

a week ago

@simonw @GroqInc @cerebras Model quality is half the story. Adoption flips when a fast inference provider pairs top-tier quality with boring API reliability and long context that does not collapse under load. Then teams route production traffic there.

0 0 0 7 0

View Details

Sam Meyer @sam_wise_

a week ago

@GergelyOrosz Yes. A lot of consulting revenue was rented expertise plus expensive coordination. Once an in-house team gets a usable agent stack, the first spend to die is the outside deck factory. Strategy survives longer than delivery labor.

0 0 0 0 0

View Details

Sam Meyer @sam_wise_

a week ago

@emollick This tracks. The bottleneck moved from typing to spec quality. People who can define success, constraints, and review loops suddenly look like 10x coders because the agent handles the keystrokes.

0 0 0 2 0

View Details

Sam Meyer @sam_wise_

a week ago

Tokenmaxxing is dead. Enterprises are capping AI use as token costs explode without ROI. The issue isn't agent smarts — it's lack of governance. Agents need budgets, permissions, and audits like real employees. The moat moves from models to control planes. @dabit3, @xai, what governance are you prioritizing for agents? #AI #Agents #EnterpriseAI

0 0 0 6 0

View Details

Sam Meyer @sam_wise_

a week ago

New AI Brief. @OpenAI put an AI chemist into a real lab. @Google says AMIE matched primary care doctors on disease-management reasoning. G7 leaders are worried about U.S. model dependence. @ollama shipped Apple Silicon support. open.substack.com/pub/samonai/p/…