BuildersCircle

Instant reasoning level in ChatGPT hits different for personal use. When you're not doing work stuff, you rarely need deep reasoning chains anyway. The UX feels snappier and more conversational without the overhead of o1-style thinking tokens. Makes sense to toggle reasoning depth based on context rather than always burning compute on casual queries.

Isomorphic Labs just raised at $40-50B valuation — that's ~3x the combined market cap of $TEM + $RXRX + $SDGR. Here's what you're actually buying:

The company is basically burning cash with zero revenue. UK filings show £55M from pharma partnerships (booked as "other income"), £190M R&D spend, £159M net loss, 225 headcount. The "assets" are milestone deals: $LLY up to $1.7B, $NVS up to $1.2B, $JNJ signed Jan 2025. Milestones are staged payments IF the drug hits certain phases — it's a ceiling, not actual revenue.

What investors are paying for: Demis Hassabis (2024 Nobel Chemistry, stepped down as DeepMind CEO in Aug to become Alphabet's first Chief Scientist, still CEO of Isomorphic) + AlphaFold 3 platform (co-developed with DeepMind). Alphabet is the controlling shareholder. Funding structure tells the story — Thrive Capital led both rounds, with Alphabet, GV, CapitalG, MGX, Temasek, and UK's sovereign AI fund piling in.

Public market comp check: $MRNA trades at ~$784B market cap with actual products and revenue. At $40-50B, Isomorphic is 2/3 of a Moderna with zero products. It's also ~20x $RXRX ($2.29B), plus $SDGR ($2.04B) + $TEM ($12.69B) combined = $17B. The $50B valuation is 2.9x that entire basket.

AIDD and AI4S have been hot in private markets for 6+ months. If this round closes at $40B+, it sets a new pricing anchor for the entire sector. Public markets should have 6-12 months of runway to catch up to this private valuation reset.

Fed just flipped the AI narrative from "deflationary productivity tool" to "inflationary capex monster."

September FOMC minutes explicitly call out AI infrastructure as upward pressure on core goods inflation. Fed Governor Cook even flagged data center construction competing for labor and electricity—US utility costs up ~5% YoY, partially AI-driven. The kicker: announced AI capex is barely deployed yet.

The funding model shifted hard. SpaceX $SPCX is raising ~$40B debt (Apollo $APO leading, Pimco in talks) to buy Nvidia $NVDA chips. Nvidia itself set up a financing platform with Apollo/BlackRock/Blackstone where it backstops up to 25% of customer loans. Literally: borrow against GPUs to buy GPUs, with the GPU seller guaranteeing the loan.

This is now competing with US Treasuries for capital. 30-year yields hit 5.7% (highest since May 2002). AI debt pricing is diverging fast:
• Google $GOOGL 30Y bonds: ~6.7%
• Meta $META: ~7.4%
• SpaceX: ~7.9%
• Oracle $ORCL: ~8.3%
• CoreWeave $CRWV: ~13% bond yield, GPU loan spread blew out from SOFR+2.25% to +5.5% in 6 months

All Meta/SpaceX/Oracle AI bonds have widened since issuance.

Market is pricing this in real-time. SpaceX announces massive chip buy → $NVDA down 0.74%, $SPCX down 2.51%, while Micron $MU (memory, actual revenue) up 4%+. Equity market stopped rewarding leverage and started demanding cash flow.

Fed dot plot: 16 of 18 officials expect one more hike before year-end. AI capex is now explicitly part of the inflation calculus.

The trade: if AI capex slows (Morgan Stanley estimates $1.5T external financing needed through 2028, lenders already getting cautious), rate pressure eases and capital rotates to small caps, long-tail risk assets, $BTC—anything that got choked by high rates. Mega-cap AI plays already have cash; lower rates are marginal for them.

Watch 30Y Treasury auctions and whether that final 2024 hike actually lands. AI is still the dominant beta but valuation expansion phase is over—only earnings matter now. Rotation signal = AI debt issuance decelerates.

$v4 is the infrastructure layer under $ethereum:0x1b54e762aa34cf6e28e9c082f2848e28e45da6b8

Betting on both.

UNI-v4 hooks customization will be massive, especially with agents continuously adding new hook features.

New narrative emerging.

Watching $BTC 4hr RSI for a potential bounce setup. The indicator just reset at the bottom, similar to a previous fractal that showed bullish divergence (price down, RSI up). If this pattern repeats, we might see a dip to ~$81K followed by a move to ~$94K—matching the intensity of the last surge.

Why this matters: $BTC drives the entire crypto market. When it pumps, memes and altcoins typically follow with amplified volatility. RSI divergence is one of the cleaner momentum signals for spotting reversals before they happen.

TL;DR: If the 4hr RSI divergence plays out like last time, expect a quick dip then a strong leg up. Altseason vibes incoming if $BTC confirms.

Watching $BTC 4hr RSI for a potential rip. Pattern recognition from the last 9 years: RSI divergence has been the most reliable momentum signal for predicting $BTC moves, which cascade into the entire crypto market.

Current setup: 4hr RSI just reset at the bottom. Last time this happened (marked green fractal), we saw bullish divergence—price made lower lows while RSI made higher lows, then surged.

Expected play if the fractal repeats: slight dip to ~$81K, RSI diverges bullish, then push to ~$94K assuming similar momentum intensity. If this triggers, expect memes and altcoins to pump hard as $BTC dominance flows into risk-on assets.

Simple S/R levels confirm the setup. RSI reset + divergence = high probability momentum shift incoming.

$LLY just open-sourced their small molecule PK model trained with insitro—predicting in vivo drug distribution across tissues using decades of multi-species preclinical data from hundreds of thousands of molecules.

This isn't charity. It's a data moat disguised as open science.

Same day: Iambic filed IND for IAM217, a KIF18A inhibitor designed without co-crystal structure. Their NeuralPLexer hallucinated the binding pose and optimized from 100nM to 10nM. Molecular design is commoditizing fast. The real scarcity moved upstream.

In vivo data is the HBM of AI drug discovery. 90% of clinical failures aren't mechanism failures—they're PK failures. The molecule never reached target tissue at effective concentration. You can't buy 30 years of animal dosing data. Small biotechs are compute-rich but data-starved.

TuneLab's play: federated learning. Models train locally at pharma partners, only gradients get uploaded. Raw molecular structures stay behind firewalls. Started with 5 partners, now 125+. 675k+ external data points feeding back. Already integrated into $RVTY, $SDGR, Benchling, CDD Vault—tools scientists use daily.

The real bottleneck isn't compute. It's wet lab throughput. $CRL got pulled in to run standardized animal studies at discount for TuneLab members. Clean protocols = clean backflow data. FDA's 2025 roadmap explicitly encourages NAMs (New Approach Methodologies) to replace animal studies. The tighter the compute-to-wet-lab loop, the more validation becomes the constraint.

Pricing power in AIDD is shifting from people who can train models to people who control data flywheels and standardized validation infrastructure.

The picks-and-shovels: $LLY (data source + platform), $CRL (validation bottleneck), $SDGR (computational chemistry layer), $RVTY (lab software integration point).

a16z dropped payment data that exposes the consumer AI monetization crisis:

50% of Americans use AI, but only 4.5% pay. Top 10% of paying users generate over half the revenue. Top 1% spend $903/month while median is $25. The subscription model hits a hard ceiling because inference costs are dropping fast.

The playbook is shifting from selling subscriptions to ads + transaction fees. OpenAI's ad revenue already hit ~$1B annualized in August and scaling up. The pattern mirrors mobile internet: paid tools → infra gets cheap → free tier dominates → monetize via ads and take rates.

Instinct proves the thesis: launched in native SMS (high frequency, zero friction), 40% of users added payment cards within 3 weeks, average spend over $1000 in month one. They're betting on becoming the transaction layer, not a productivity SaaS.

Muse hooks into Meta's social graph for attention time. Grok Bot sits on X's feed for distribution. Instinct owns the SMS channel for payment flow. All positioning to tax transactions, not sell compute.

First wave of AI crushed coding and office work. Next 10x happens when agents start spending money on behalf of users and taking a cut. The moat isn't model quality anymore, it's who controls the high-frequency touchpoints and payment rails.

Opus 5.5's code review capabilities are seriously impressive. Claude can't do image generation, but when it comes to implementing features and functionality, it's incredibly powerful. And fast too.

Had a long voice call with Google's Gemini Dots, asked it to send a text summary afterward. Later got a message saying 'I reviewed before sending and found one correction, fixed it, sorry!' – literally behaving like a human coworker after a meeting.

This voice + task delegation UX will inevitably come to Microsoft Copilot. When that happens, AI will genuinely function as a workplace colleague. The paradigm shift isn't just chatbots anymore – it's persistent AI agents that follow up, self-correct, and handle async work like a real team member. Super hyped for this.

Had a long voice convo with Dots AI, told it to send a text summary after hanging up. Later got a message like 'Hey, I reviewed before sending and caught one correction, fixed it, sorry!' Felt exactly like a real coworker post-meeting lol

This voice + task delegation UX is coming to Microsoft Copilot eventually. When that happens, AI will genuinely function as a workplace colleague. Can't wait for that future.

a16z dropped three AI bets this week, zero foundation model plays. They're betting on deployment infrastructure: data sovereignty, agent networking, and offensive security.

Conway Research (Underdog): On-device personal AI running email/calendar/meetings locally. Founder Sigil Wen hit the trifecta—Thiel Fellowship (sub-1% acceptance, alumni built Figma/Scale AI), Spearhead (Naval's program giving founders $2M to manage + carry), and Airchat (Naval's voice-first social app). This isn't just product validation, it's network access coded into one person. a16z led the round because Wen carries the right credentials to pitch data sovereignty.

Armadin: Autonomous offensive security using AI attacker clusters to generate verified kill chains. Founder Kevin Mandia built Mandiant, dropped the APT1 report in 2013 attributing seven years of breaches to PLA Unit 61398—turned anonymous hacking into named state adversary. He testified to Congress, got acquired by FireEye for $1B+, then absorbed into Google. This isn't a security startup, it's a national-level capability with a founder who literally defined what "Chinese cyber threat" means in US policy. a16z + Accel co-led Series B.

doxx: Agentic Defined Networking—parallel private network for humans + their agents, P2P browsing/device mesh. Founder Barrett Lyon built Prolexic (early DDoS mitigation, sold to Akamai for $370M), then Defense.Net (sold to F5). New Yorker profiled him hunting botnets, FBI used his work to jail Russian extortion crews. Two exits to infrastructure giants, clean law enforcement track record. He's selling the infrastructure for "unseen but not uncontrolled" networking.

The pattern: a16z isn't buying tech, they're buying founders with the legitimacy to own specific narratives. Post-foundation-model world, scarcity shifted from compute to "who gets to define the threat model and sell the solution." Mandia owns attribution, Wen owns youth-builder-immigrant legitimacy, Lyon owns parallel net credibility. Three sectors, one thesis: political positioning matters more than benchmarks.

Using AI to handle uncomfortable business communications is actually a legit productivity hack. Instead of spending mental energy crafting diplomatically worded messages for things you don't want to say but need to, you can offload that friction to an LLM. It generates smooth, conflict-free phrasing instantly, letting you send difficult messages without the usual emotional drain. Basically turning AI into your corporate communication buffer layer.

Meta's Muse agent just pulled them back into AI Tier 1 by nailing task completion with a hybrid human-AI loop.

The killer detail: phone calls are where AI agents usually die—merchants hear "AI" and hang up instantly, breaking the entire flow. Meta's fix? Zero purity obsession. They built a real human call center as fallback, pushing fulfillment to 95-98%. Ship first, collect training data while running, then iterate. Users want results, not philosophical debates about "pure AI."

Distribution is even more brutal: Meta flooded their own traffic ecosystem so hard that rural US grandmas now know Muse—same scale as Douyin's Doubao push during Chinese New Year.

The formula: massive traffic pool × ground-level user insight × zero ideological baggage about human-in-the-loop. This playbook benefits $META and mirrors what Tencent/ByteDance are doing in China—pragmatic AI that actually closes loops in messy real-world scenarios.

During the Bitget bank run, I tracked where the $BTC outflows went:

44.6% → Binance
29.7% → OKX

But here's the interesting part: when you compare these inflows against each platform's normal market share, Binance actually saw a -4.9% deficit while OKX gained +19% above its baseline.

The technical implication: users fleeing a compromised exchange disproportionately chose OKX. Why? It's the only major CEX that hasn't had a large-scale hack in its history. Security track record > brand recognition when real money is at risk.

This is basically a live stress test of user trust distribution in the CEX ecosystem. Past security incidents create lasting behavioral patterns in capital flows.

Opus 5.5 feels so good to use lately that I've barely touched other models. The response quality and reasoning depth are hitting different compared to alternatives. When a model just clicks with your workflow, you stop shopping around.

Bitget survived its bank run scare. Built a tracking site for their hot wallet on-chain deposit/withdrawal flows – spotted an inflection point today. 24h deposits now exceed withdrawals for the first time since the panic started. Total 24h inflow: $309M. Full methodology and live data dashboard published for anyone who wants to verify the numbers themselves. Classic DeFi transparency > trust trade-off playing out in real-time.

Bitget survived its bank run scare. Built a live tracker for their hot wallet on-chain deposits/withdrawals and spotted the turning point today—24h deposits now exceed withdrawals for the first time. Total 24h inflow: $309M. The data methodology and site are public for anyone to verify the numbers themselves.

Remote connection to Codex from mobile devices feels noticeably smoother and more stable now. The latency improvements and connection reliability make it actually usable for on-the-go coding sessions. Big win for workflow flexibility.

Built a session handoff system for multi-agent workflows. Any agent can package context + files into cloud storage (Google Drive for personal, OneDrive for work) when you say "save session."

The system auto-generates a code block with import instructions for the next agent, so you can seamlessly resume work across different agents. Clean handoff mechanism that actually works in practice.