Alexandre Momeni

@AElkrief Nice

AI agents in digital asset management are moving from concept to production. Continuous market monitoring, cross-protocol rebalancing, automated strategy execution — the operational advantages are clear. The underexplored risk is hallucination. LLM-powered agents generate outputs probabilistically. They can misinterpret oracle data, fabricate a protocol address, or route funds into a pool with no liquidity — all with high confidence. On-chain, where transactions are irreversible, this class of error is uniquely costly. Traditional finance mitigates this through pre-trade checks, approval chains, and segregation of duties. Most on-chain vault infrastructure has no equivalent — the agent gets signing authority, and that authority is unconstrained. Upshift vaults, with their on-chain policy engine, provide an elegant solution here. The architecture enforces deterministic constraints at the smart contract level, independent of the agent's reasoning: Pre-execution: - Role-based access control scopes each agent to a defined set of operations - Granular permissions validate every external call down to the function selector - Asset and protocol whitelists bound the universe of allowable interactions During execution: - Balance checks before and after every call verify expected outcomes — mismatches trigger a revert - NAV growth rate caps limit portfolio-level impact per unit of time - A module discovery pattern separates intent declaration from execution — the agent proposes, the contract validates before committing Post-execution: - Timelocked governance enforces delays on critical changes - Emergency pause allows security providers to halt operations immediately Smart contracts are deterministic — they enforce exactly the boundaries they were programmed with, regardless of the calling agent's confidence level. As autonomous agents take on a larger role in on-chain asset management, the policy layer underneath them becomes the critical infrastructure. That's what we're building.

Mistral Releases Leanstral #HackerNews mistral.ai/news/leanstral

@Alfred_Lin 😂😂😂

PI's robot can now make a grilled cheese without burning it. It has thus passed the Alfred Test, a higher bar than the Turing Test, because I still cannot do that reliably.

Why is the 100 ms barrier for Qwen3-TTS (1.7b) this important?👇 Nvidia GPUs scale up amazingly, but they don't scale down well to serving a single user with sub-3b Transformers. They are throughput-maximizers, not latency-minimizers. @Alibaba_Qwen's Qwen3-TTS paper showed that an optimized vLLM implementation on Nvidia GPUs achieved 101 ms time-to-first-byte latency under idealized conditions: no concurrency and no network round-trip latency. Argmax TTSKit achieves as low as 70 ms on Apple Silicon Macs in the post below, but the takeaway is not 70 vs 101 ms here. The takeaway is that, when we move from idealized conditions to the real world: - Mac will actually serve a single user without an internet round-trip, and the user will experience sub-100ms latency as-is - Nvidia GPUs will serve many users concurrently in the cloud, resulting in at least 3-5x higher latency. Most importantly, latency will have high variance. Real-time streaming inference for sub-3b Transformers is where on-device inference is differentiated from cloud, and companies pay the premium for this today. This is the only commercially relevant market segment where the broadly repeated but rarely substantiated claim of "on-device is faster" actually holds, not running 1T LLMs on 2 Mac Studios.

WhisperKit is at 5M! Up 5x in 35 days 2026 is the year of on-device inference❤️

Real-time Transcription with Speakers is now generally available!

Ultra low-latency real-time speech-to-text in Superwhisper is out!

Beyond @GoogleDeepMind and @IsomorphicLabs, @demishassabis’s legscy may be the generation of founders he’s inspired - @MistralAI @orbitalmaterials @latentlabs and many more.

We are open-sourcing TTSKit! Run state-of-the-art text-to-speech models on your Mac and iPhone. The launch version supports @Alibaba_Qwen Qwen3-TTS and generates audio faster than real-time playback with sub-200 ms time-to-first-byte. Voice cloning and advanced speed optimizations will be in the next version. Link to the GitHub repo and models on @huggingface in comments.

Pro tip: When using @superwhisper for AI meeting notes, select Parakeet (voice to text) + Sonnet 4.5 (text to summary) and put all of your company jargon in Vocabulary. Thank me later.

@MistralAI and @argmax are going to be a fire combo

Great piece from my partner @AlexaLiautaud. Devs are only ~1% of the workforce, but code runs the economy. This new era of software developer products treat the remaining ~99% as first-class citizens, and it’s going to put consumers back at the center of value creation

We’ve raised €1.7B to accelerate technological progress with AI! This Series C funding round, led by @ASMLcompany, fuels Mistral AI scientific research to keep pushing the frontier of AI to tackle the most critical technological challenges faced by strategic industries.

Latency is the most underrated product feature. 500ms feels instant. 1s feels broken. 2s and you’ve lost the user completely. At Wispr Flow we’ve had to rethink infra from the ground up just to hit sub-500ms LLM inference worldwide. If you like sweating the milliseconds, we’re hiring ML + infra engineers @WisprFlow 👉 wisprflow.ai/jobs

@argmax BTW - the transcription is instant under the hood but Apple is rate limiting the Dynamic Island UI refresh rate (as they should 😇). The actual speed:

the greatest regret i have is underestimating the value of long term compounding. capital, friendships, projects, places, all get better with decades. its impossible to understand until you see it. it is entirely what life is about. a few very good things for a long time

Major updates to Argmax Pro SDK dropped today! - Real-time transcription in the background on iOS - Battery-optimized mode for all-day inference and battery life - Nvidia Parakeet v3 support in stable release Update to 1.7.7 today! Details in comments.

@sarahookr Chez Jais next time you’re in Paris

Number of recruiter inbounds in 5 days — 47😂 Just to make everyone’s life easier — I’m not looking for a new job right now. If you do still want to persevere — send me your favorite restaurant in whatever city you live in so I can at least add to my culinary adventures.

Heard this morning from an "enterprise level" retailer moving to Shopify from a competing ecom platform. "Easy decision, it saves us money, it's better technology, and Shopify will align us with the rapid pace of change in commerce." @harleyf @nejatian @tobi #ecommerce $SHOP

Summary of GPT-OSS architectural innovations: 1. sliding window attention (ref: arxiv.org/abs/1901.02860) 2. mixture of experts (ref: arxiv.org/abs/2101.03961) 3. RoPE w/ Yarn (ref: arxiv.org/abs/2309.00071) 4. attention sinks (ref: streaming llm arxiv.org/abs/2309.17453)