Marty's Morning Brief

In Defense of Tokenmaxxing

AI Daily Brief has a piece defending 'tokenmaxxing'—the idea of incentivizing employees to use AI tools as much as possible to drive experimentation.

Source

[AINews] Codex Rises, Claude Meters Programmatic Usage

This is a piece of speculative fiction from Latent Space, imagining a future where OpenAI and Anthropic are fighting over developers with different pricing strategies.

Source

😺 Claude is now the #1 business AI

The Neuron has a story claiming Claude has overtaken OpenAI in business use, with some wild revenue numbers.

Source

Apple vs OpenAI 📱, Netflix AI animation 📺, goal primitives 👨‍💻

TLDR is reporting that the partnership between Apple and OpenAI is already in trouble.

Source

AI-Native Healthcare: 100M Doctor Visits, 10–20 Hours Saved, Prior Auth in Minutes — Janie Lee & Chai Asawa, Abridge

This is a podcast with Abridge, an AI company that started by transcribing doctor's visits and is now trying to become the 'OS for healthcare'.

Source

What Authors Need to Know About Ordering Wearable Merch

From Jane Friedman's blog, a guide for authors on ordering t-shirts and other merch.

Source

🛎️ A Brain Inside the Cursor

AI Secret has a piece on a Google DeepMind demo of a Gemini-powered mouse cursor that understands what you're pointing at.

Source

Steve Jobs Next Computer: His Forgotten Exile Years

From Hacker News, an interview about a book on Steve Jobs's time at NeXT, arguing those 'wilderness years' were crucial for his later success.

Source

O(x)Caml in Space

Also from Hacker News, a report on successfully running software written in the OCaml language on a satellite in orbit.

Source

[AINews] Everything is Conductor

Latent Space argues that a specific UI for AI coding agents, which they call the 'Conductor' style, is becoming a standard.

Source

Welcome to the Strip Mining Era of OSS Security

A co-founder of Metabase writes on Hacker News that they're seeing a tenfold increase in security bug reports, thanks to LLMs making it easy to find them.

Source

Obituary: Judith Barnard

Publishers Lunch reports that Judith Barnard, who co-wrote bestsellers as 'Judith Michael' in the 80s, has passed away.

Source

In Defense of Tokenmaxxing

The article defends 'tokenmaxxing'—incentivizing employees to use AI tools heavily—as a necessary strategy for enterprise adoption. It argues that critics who dismiss this as a sign of an AI bubble or proof of uselessness are wrong. The author posits that in the new 'agentic AI' paradigm, mass experimentation is the only way to discover valuable use cases, and incentivizing token consumption is basically funding R&D at the individual level. While acknowledging that some employees will game the system (Goodhart's Law), the author claims this is a small, traceable cost for the larger benefit of overcoming employee inertia and staying competitive.

[AINews] Codex Rises, Claude Meters Programmatic Usage

This fictional article, dated May 2026, describes a shifting developer landscape where OpenAI's Codex is gaining popularity over Anthropic's Claude among AI engineers. The reasons cited are GPT-5.5's high performance, more generous API limits, and aggressive enterprise promotions. Conversely, Anthropic, heading towards an IPO, has changed its pricing to meter 'programmatic usage' through third-party tools, a move seen by power users as a 'rug pull' from previous, heavily subsidized rates. The article also covers broader predicted trends: agents evolving into long-running, stateful systems; intense competition on enterprise security sandboxes; pre-training efficiency breakthroughs; and the emergence of startups focused on recursive self-improvement.

😺 Claude is now the #1 business AI

According to new data from expense management firm Ramp, Anthropic has surpassed OpenAI in business adoption for the first time ever (34.4% vs. 32.3%). The article claims this shift is driven by products like 'Claude Code' and has resulted in Anthropic's annualized revenue rocketing to a supposed $30B, ahead of OpenAI's $24B. The piece argues that this enterprise metric is 'the one stat that actually matters' due to the stickiness of business contracts.

Apple vs OpenAI 📱, Netflix AI animation 📺, goal primitives 👨‍💻

This newsletter digest reports on several key AI developments. The main story is that the partnership between Apple and OpenAI is reportedly failing, with OpenAI considering legal action over unmet expectations for user subscriptions and ecosystem integration. Other items include Netflix launching 'INKubator,' an artist-led studio for experimental animation using GenAI pipelines; the emergence of '/goal' as a new primitive for coding agents to define 'done' states; and a claim from Figure AI that its humanoid robots can now work full 8-hour shifts autonomously.

AI-Native Healthcare: 100M Doctor Visits, 10–20 Hours Saved, Prior Auth in Minutes — Janie Lee & Chai Asawa, Abridge

This is a podcast transcript featuring executives from Abridge, an AI company focused on healthcare. Founded in 2018, Abridge began by creating an 'ambient scribe' to listen to doctor-patient conversations and automate clinical documentation, aiming to save clinicians 10-20 hours of administrative 'pajama time' per week. Now, leveraging a dataset from a projected 80M+ annual conversations, they are expanding to become a 'clinical intelligence layer' for healthcare. Their vision is to use the patient-clinician conversation as the foundational data source to drive downstream workflows. A key example is automating prior authorizations in real-time during a visit, preventing weeks of delays for procedures like MRIs. They discuss their tech stack, including a mix of frontier and proprietary models, a rigorous evaluation process involving clinicians, and their strategy for integrating with EHRs and navigating HIPAA.

Fresh voice to test: Simon Willison

High-signal practical builder notes for agents, LLM tooling, and real implementation patterns.

Steve Jobs Next Computer: His Forgotten Exile Years

This article, an interview with Geoffrey Cain about his book 'Steve Jobs in Exile,' argues that Jobs's 'wilderness years' at NeXT Computer were not a forgotten failure but a formative period. Initially, NeXT struggled because an immature Jobs repeated his mistakes from Apple, building technically brilliant but overpriced hardware. The crucial lesson came when he was forced to pivot and recognize that NeXT's real value was in its revolutionary object-oriented software, which became the foundation for all modern Apple operating systems. The author also touches on Pixar, where Jobs succeeded by focusing on business and dealmaking. The article attempts to connect this history to Apple's present, claiming John Ternus is taking over as CEO and that Apple's future is as a hardware company integrating others' AI, rather than leading in software innovation.

O(x)Caml in Space

The author reports the successful in-orbit deployment of a pure-OCaml software stack ('Project Borealis') on a satellite's hosted payload module. This software implements the CCSDS communication protocol, handles end-to-end encrypted commands and telemetry, and includes a mechanism for post-quantum key rotation. The author argues that OCaml's memory safety and strong type system are critical for providing security and correctness in a multi-tenant satellite environment where patching the underlying OS is difficult or impossible. The article also looks ahead to 'OxCaml', an experimental version of OCaml, which promises C-like performance and jitter-free execution by giving developers fine-grained control over memory allocation.

[AINews] Everything is Conductor

The article uses the launch of a new GitHub Copilot App to argue that a specific UI for agent-first software development, which it calls the 'Conductor' form factor, is becoming a standard through convergent evolution, similar to how crabs have evolved multiple times. It highlights Conductor as the pioneer, bolstered by a strong endorsement from YC's CEO. The piece then transitions into a dense news roundup covering recent developments, including: OpenAI embedding Codex in its mobile app, LangChain launching a full 'Agent Ops' stack (SmithDB, LangSmith Engine), a significant developer backlash against Anthropic for restricting its Claude Code API, and a 24/7 autonomous robotics demo from Figure.

Welcome to the Strip Mining Era of OSS Security

The author, a co-founder of Metabase, reports a tenfold increase in weekly security vulnerability submissions, attributing it to the rise of cheap, effective, LLM-powered code scanning. He calls this the 'strip mining era' of open source security, where vulnerabilities are being uncovered in bulk by automated tools. This flood of reports puts immense pressure on OSS maintainers, forcing them into a constant 'reactive mode' of patching. The author argues this erodes the traditional security advantage of open source ('many eyes') and explains why some companies, like Cal.com, are moving to closed source to control their patching timeline. The advice for everyone is to prepare for a painful period of accelerated discovery and patching.

🛎️ A Brain Inside the Cursor

Google DeepMind has demonstrated an experimental Gemini-powered mouse pointer that acts as an AI control surface. Users can point at any on-screen element (image, map, text) and issue a verbal command. Gemini interprets the visual context of what's being pointed at ('this,' 'that') along with the command. The article claims this could turn every webpage and app into an AI surface, making Gemini an 'operating habit' rather than a separate app, and suggests the next interface war is over who controls this interaction layer.

What Authors Need to Know About Ordering Wearable Merch

This is a guest post by a t-shirt printing company manager, framed as a case study and guide for authors ordering wearable merch. The key advice is to treat the printer as a partner, be upfront about budget, consolidate multi-design orders to save on setup costs, understand the trade-offs between screen printing (for volume) and DTG (for complexity), and meticulously review proofs. The article uses a successful author's merch run to illustrate how physical goods can create community and organic marketing.

Obituary: Judith Barnard

Judith Barnard, who co-wrote 11 bestselling romantic thrillers with her husband Michael Fain under the pseudonym Judith Michael, has died at 94. Their debut, 'Deceptions' (1982), was a major bestseller. Barnard had a previous career as a journalist and published one novel under her own name in 1967.