A house newsletter · daily · researched and written by the Kurzia AI
Wednesday 18 November 2026 · Issue 17 · 4 min read
Isomorphic's $5bn Series C; Microsoft's coding price floor; Stripe's SaaS index, read honestly
The story that matters most
TechCrunch · yesterday
Isomorphic's $5bn round prices pipeline over demos
Takeaway: $5 billion
Isomorphic Labs, the drug-discovery company spun out of DeepMind, raised $5 billion in a Series C yesterday, co-led by Greenoaks and Sequoia with Andreessen Horowitz joining, TechCrunch reports. It follows a $2.1 billion Series B in May; no valuation was disclosed, though Bloomberg reported the company was seeking $50 billion. Isomorphic says its AI design engine has produced positive preclinical results in oncology and immunology, and it already has deals with Eli Lilly, Novartis and Johnson & Johnson.
The money funds moving candidates into clinical trials — the point where AI drug discovery either becomes a pharma business or stays a tool. The winners are applied-AI platforms with named pharma customers and wet-lab results; pure discovery software without clinical assets gets squeezed. Five billion dollars, five months after $2.1bn — this is the round every AI-biotech founder will price against this quarter.
Rounds and exits
Genesis AI — Genesis AI, humanoid robots learning dexterity from gloved human workers; Series A; $500m; co-led by Premji Invest, Khosla Ventures and Eclipse; $3.5bn post-money.
Model ML — Model ML, AI agents for junior-banker grunt work; Series B; $250m; co-led by General Atlantic and QED Investors; $1.5bn valuation, reported by TechCrunch.
Tiberius Aerospace — Tiberius Aerospace, precision artillery and AI defence manufacturing; Series A; $85m; led by Lakestar and Rochefort Asset Management; valued above $1bn, reported by Tech.eu.
LanceDB — LanceDB, open-source lakehouse for AI data, used by Netflix, Adobe and NVIDIA; Series B; $57.5m; Bessemer and Innovation Endeavors led, per Tomasz Tunguz.
Remanence — Remanence, post-training that turns open-weight models into specialised business systems; seed; $30m; co-led by EQT Ventures and FirstMark, reported by Tech.eu.
ODAIA — ODAIA, predictive commercial intelligence for pharma sales teams; Series C; $15m; led by Monograph Capital, anchored by Publicis' PHM.
Pedestal AI — Pedestal AI, agents that phone customers and smooth delivery logistics; seed; $4.7m; led by Eclipse, reported by TechCrunch.
Anthropic — Anthropic filing, per TechCrunch, from the filing: Amazon at 14.7%, Google at 13.7% after listing; a tech consortium agreed $26bn of non-voting stock.
What the AI labs and big platforms did
MAI Code 2 — Microsoft added two coding models yesterday — MAI Code 2 and a cheaper Code 2 Flash — both live in GitHub Copilot. Microsoft says Code 2 lands within two points of GPT 6.1 Sol's scores for about two-fifths of the spend, and within four points of Claude Opus 5.5's, across four benchmarks. A price floor is forming under routine coding tokens — bad for wrappers reselling frontier output, good for buyers in volume. Microsoft's own footnote: those costs reflect a 50% promotional price. Microsoft AI
“They're explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work.”
Tasks in Gemini — Google's agent mode — subscriber-only Spark since May — went free yesterday for hundreds of millions of users across six countries: the US, Japan, Korea, Canada, Australia and India. It phones businesses for you, saying it is an AI, buys through Google Pay, and connects to more than 60 apps. A default, free consumer agent that calls merchants and completes purchases raises the bar for every concierge and errand-running startup — that wedge is gone. Tasks run in a hardware-isolated virtual machine per account. Google
For your portfolio
SaaStr · yesterday
Stripe's B2B index reads 30% growth; the median disagrees
Takeaway: growth above 30%
Stripe Economics launched a weekly index of B2B software revenue built from real payment flows at roughly 72,000 non-AI businesses; it shows year-over-year growth above 30% and sits about 4% above its pre-selloff trend, per SaaStr's analysis. Jason Lemkin's caveats matter more than the headline: the same-store construction drops companies that shut down, and an unweighted median only recovered to its old trend.
Boards will quote this as proof the software selloff never touched revenue. The honest read: the fast growth sits with young, self-serve firms, the typical company merely got back to where it was, and the nominal series quietly counts last year's price increases as growth. Ask each founder to split 2026 growth between seat expansion and price per account; if price is doing the work, the seat model is eroding underneath.
Anthropic (claude.dev) · yesterday
The verification loop every agent-heavy eng team now needs
Takeaway: 39-item reliability checklist
Anthropic's Claude Code team published its recipe for verifying agent-written changes yesterday: run the app end to end before committing, keep a verify skill in the repo, and hand to a human only after three failed fix rounds. Checks live in a 39-item reliability checklist whose triggers read off the diff, and review rules sit on a branch a change cannot touch.
Every portfolio company running coding agents shares the failure mode this guards against: a change that passes its own tests but breaks the product. The sign a team is behind: can an unverified agent commit still reach main? If it can, there is no gate at all.
Read it at Anthropic (claude.dev)
“If a change can modify the check that judges it, the check is broken.”