swyx

Jamin Ball × swyx

Anthropic’s S-1 reports 85% gross margins, 21% latest-quarter free cash flow margins and 8% GAAP operating margins; cash of $134 billion compares with $141 billion raised.Serving-compute spend rose 2.3x versus 14x revenue, implying roughly 7x efficiency gains, but margins may turn negative as investment accelerates; lenders’ appetite for concentrated exposure will test Ball’s illustrative roughly $100 billion annual commitment schedule.

Zach Lloyd × swyx

Code-defined factories give each product surface a versioned workflow, shared traces, and replaceable compute, inference, and harnesses.Warp reports PR cost fell from about $80 to roughly $15–$20, while throughput rose 40% last month without higher revert rates, citing cheaper triage and less redundant agent work.Replay validation of agent-proposed changes remains unfinished; Lloyd places the likely adoption floor at 5–10 engineers.

Basil Chatha × swyx × Jacob Lauritzen

Legora’s edge is deep ownership of legal workflows, not access to frontier models.Client confidentiality blocks usage-trace access, so about three engineers, three lawyers and 100 legal engineers reconstruct evaluations; models handle roughly 60% of work well, but regressions persist.As routine legal fees face pressure, swyx argues clients will still choose Legora for a $400 million M&A transaction.

Joseph Kim × swyx × Richard

Bundled data differentiates ChatGPT for Financial Services: Kim says some major MCP connectors had a “40%-plus error rate,” making data customers’ top positive callout despite duplicate-license risk.Codex supplies about 80%, but an unchanged finance launch “would fail”; “95% of what we built” is reproducible by others.Watch human review and banking evals as autonomy rises in ’27 and ’28.

swyx × Katelyn Lesse × Angela Jiang

S-1 figures show revenue per employee rising from $1.4 million to $6.6 million and a first 19% GAAP-profitable quarter.More than 50% of customers above $100,000 in trailing-12-month revenue use Claude Platform; Managed Agents keeps credentials beyond agents’ view.A Claude-assisted Rust port cut CPU use 40–50% and first-token latency 10–20%, while CPU scarcity and personalization remain constraints.

swyx × Paul Copplestone × Ant Wilson

Turso and Spark target billions of databases with a lower-cost start and upgrades to Postgres and Multigres.Founders say agent-driven activation is almost 100% and paid conversion exceeds traditional developers, while salespeople earn on incremental growth: 125% monthly versus 25% for nonresponders.Enterprise expansion remains greenfield-focused because production-database migration risks downtime and split-brain data.

Anekha Sokhal × swyx

Moshi uses an internally built recommendation system to surface second- and third-order market effects a news stream may miss.Its graph combines futures, fundamentals, news, X, podcasts, and private data; Sokhal says answers that take days can arrive in 30 seconds.Yet swyx questions trading on noisy, potentially manipulated X data, citing about $300,000 monthly for Balyasny's X Firehose; Moshi scores confidence, but any opportunity remains unknown.

Alex Lieberman × CJ Hess × swyx

Among about 100 enterprises 10X talks to, 90% have rolled out Claude or Codex, but only about 5% have non-engineering agents in production, revealing the last-mile gap.CJ estimates output roughly doubled from December’s “Opus 4.5 moment” to Opus 5.5; scaling still requires process standardization and permissioning.

swyx × Jason Ma × Vibhu

Multi-site deployments folding thousands of napkins daily at Din Tai Fung and laundry for Best Western offer commercial proof beyond pilots; Din Tai Fung has signed a much bigger contract.Jason calls Dyna-2 the first world action model trained on more than 1 million hours of data, but missing sensing modalities remain a potential constraint on broader generality.

swyx × Brandon × Liam Fedus × Ekin Dogus Cubuk

Periodic’s differentiated asset is timestamped lineage across failed experiments, computations, instrument data, code, discussions, and successes, enabling models to learn scientific process rather than published endpoints.Its AI-simulation-experiment loop targets automated characterization, with secure on-site semiconductor deployment as the near-term wedge, while the longer-term “matter compiler” faces unknown physical variables and difficulty discovering superconductors without a complete theory.