It ships with a native Model Context Protocol server. An AI agent can programmatically discover and use components via a CLI flag (`--dense`). This is what 'AI-fluent' actually means. Full breakdown: https://t.co/KmASOcQh0T
Most design systems are built for developers. Meta just released Astryx, built for AI agents first. 13,000+ internal apps, now open source. The shift is architectural, not a plugin.
Key feature: `vibe-tests`. It scores AI-generated code on a 0-100 scale across correctness, accessibility, efficiency. Turns the trust paradox (46% of devs distrust AI output) into a measurable feedback loop.
software bots (Otter, Krisp) can't exist in a doctor's office or a factory floor. Hardware competitors like the Omi use squeeze gestures that can fail. The NotePin S's clicky button is simple, but it's the entire thesis. Full breakdown: https://t.co/R64zNlOpat
The problem with recording a conversation was never capture—it's retrieval. You get 90 minutes of audio but can't find the one moment where the client said 'yes.' I tested a $179 wearable that solves this with a single, tactile button press.
The most interesting part? For neurodivergent users, especially those with ADHD, this externalizes the executive function of 'remember what's important' into a repeatable physical action. One verified buyer called it 'fully present during meetings.'
They're building a 'two-engine' system where GPUs handle the math-heavy attention layers and this new chip handles the memory-heavy stuff. On Llama 3 70B, it runs 3-5x faster than an H100 at batch size 1. Wild
https://t.co/eHoVNYI9r9
Full breakdown on the denominator effect, the AI thesis shift, and the liquidation preference trap that could wipe out your exit: https://t.co/CRR9qoS8FP
Europe's exit pipeline tells the story. Unicorns fell from 69 in 2021 to just 7 in 2023. No exits = no returns to LPs = less capital for new funds. The cycle that funds startups is frozen, and it’s choking the entire ecosystem from the top down.
The AI squeeze is brutal. Andreessen Horowitz's entire 2025 thesis is essentially a single-tech bet. If you're a SaaS or fintech founder without an AI-native layer, you're now fighting for attention in a market where the largest VC is signaling your sector isn't the priority.
The core issue isn't just 'VCs are scared.' It's structural. Higher interest rates triggered the denominator effect: pension funds' VC allocations swell on paper, forcing them to rebalance and cut new commitments. The free money era didn't just end — it rewired the math.
A founder can build a company, raise through a Series C, and exit for $20M — only to walk away with less than $2M. This is the hidden tax of stacked liquidation preferences, and it's becoming the norm in 2025. A thread on why fundraising is harder now than in 2008. 🧵