最近の記事
30The Thriving Ecosystem of Open Models
Competition is a discovery procedure. — Friedrich Hayek And developers are discovering the value of open models. OpenRou
6/2/2026
The AI Skepticism Map
With Michael Burry 1 & Leopold Aschenbrenner 2 placing heavy short trades on AI, questions about GPU depreciation, &
6/1/2026
Skill Distillation
I’ve been using state-of-the-art models to teach small models running on my computer how I work. My personal agent
5/29/2026
Security in the Age of AI Agents: Office Hours with Jonathan Jaffe
When security practitioners become engineers, the mission changes from managing people to architecting the automated pol
5/28/2026
Software After AI
The end of the software era is the beginning of the harness era. AI outmoded SaaS managed databases with fixed workflows
5/27/2026
Agent Gravity : Who's Running Your Agents
If data gravity was the most important force in the Decade of Data, agent gravity will be the same in the Decade of Agen
5/26/2026
Plastic User Interfaces
Salesforce has gone headless : a sales person can update their deal sheet without ever logging into salesforce.com throu
5/22/2026
SpaceX's Limitless Ambition : An AI Conglomerate
After 24 years as a private company, SpaceX filed its S-1 yesterday. The filing reveals an AI-era conglomerate. SpaceX h
5/21/2026
The Unsustainable Subsidy
Google’s AI triples in price each year. OpenAI’s flagship model was seemingly subsidized for a while, before
5/20/2026
Observations on Writing with AI
As I was paging through Good Writing , Anne Lamott’s new book, I wondered what AI would say about twisting cliches
5/18/2026
The First Derivative of Inference
The fastest-growing companies in AI & software are either selling AI directly or reselling inference. At worst, they
5/15/2026
What Would AI Email Cost?
In yesterday’s post (which an agent pushed in raw outline form via email!), I wrote about the future of AI email.
5/14/2026
The 6 Messages That Actually Matter
Nobody will open Gmail five times a day in five years. The average knowledge worker receives 121 emails per day. 1 That&
5/13/2026
2026 Theory GTM Survey
It’s time for the 2026 Annual Theory Go-to-Market Survey. This is a brief 25-question survey. Our goal is to under
5/12/2026
Localmaxxing
As demand for AI inference explodes, I’ll be asking a lot more of my little computer. How much more? Over the past
5/11/2026
Securing the Agentic Enterprise
Enterprises run on AI agents. So do the attackers. What does it mean to build, secure, and operate AI systems when both
5/8/2026
AI at Discount
Anthropic grew from $1B to $30B in 15 months. So why does it trade at a discount to public comparables? High-growth comp
5/7/2026
Optimizing Software Factories
What happens when a startup employee leaves on a Monday? In a twenty-person engineering team, one resignation is a 5% he
5/5/2026
When Everyone Is a Key Person in Your Company
What happens when a startup employee leaves on a Monday? In a twenty-person engineering team, one resignation is a 5% he
5/5/2026
All the AI You Need for 8 Ads per Day
One search ad every 40 minutes pays for a trillion-parameter model. One content ad every 3 minutes does the same. The ma
5/3/2026
This Week's Sign the Apocalypse Isn't Upon Us
I remember growing up reading Sports Illustrated . There was a small column called “This Week’s Sign the Apo
5/1/2026
The $112 Billion Quarter
Google Cloud grew 63% year-over-year in Q1 2026. Amazon Web Services posted 28%. Microsoft Azure hit 40%. All three are
4/30/2026
Timing Cashflows of Your Sales Commissions and Your Customers
4/30/2026
Darwinian Specialization in AI
The inference market is the fastest growing market in the world & it’s splitting up. Each modality is developi
4/29/2026
The Three Questions in AI Sales
The old sales motion asked : what’s your software budget for this category? The new motion asks customers three qu
4/28/2026
GPU Spot Prices Surge 114% in Six Weeks
NVIDIA’s latest GPU rental prices on the Ornn Compute Price Index hit $4.95 per hour this week, up from $2.31 in e
4/27/2026
Competitive Strategy in the Age of AI
Google commoditized its complements : free maps, free email, free browsers, a free mobile OS. They removed every toll bo
4/24/2026
Intelligence About Business
“BI became dashboards. And now it is re-expanding to business intelligence.” — Colin Zima, CEO & co-foun
4/23/2026
A $10 Billion Call Option
SpaceX announced a partnership with Cursor today : a $10 billion collaboration with a $60 billion acquisition option lat
4/21/2026
The Price of Precision
“As models get smarter, they can solve problems in fewer steps : less backtracking, less redundant exploration, le
4/19/2026
投資シグナル
20Market skepticism is concentrated in AI companies dependent on future capital access and demand rather than broad fatigue with AI infrastructure, with memory becoming the critical bottleneck.
Small and mid-cap AI companies with future-dependent business models face concentrated short selling, with SoundHound AI at 36.3%, C3.ai at 32.2%, and others heavily shorted.
SaaS and developer tools companies face negative sentiment with short interest at 9.5%, driven by need to prove business value post-AI investment cycle.
GPU data center businesses face significant skepticism with short interest growing 60% year-over-year, indicating market concerns about GPU depreciation and current pricing versus long-term value.
Chinese AI labs (DeepSeek, MiniMax, Alibaba, Qwen, Tencent) and emerging US labs (Arcee) are competing aggressively in open-weight models, with leadership positions shifting multiple times since 2025.
Competition in the open-model ecosystem is driving rapid innovation and continuous leaderboard shifts, with new model releases sustaining plateaus of increased token volume and developer adoption.
Open-weight models are rapidly gaining market share, reaching 69.1% of named token volume on OpenRouter as of 2025, indicating developers increasingly prefer and route production traffic to open models over closed alternatives.
The author believes institutional knowledge can be effectively encoded as procedural skill libraries that are interpretable, versionable, and dynamically swappable across different model backends based on cost.
There is increasing adoption of agent-based systems that combine local knowledge bases, skill libraries, and API integrations to create specialized operating system-like personal assistants.
Skill distillation is an emerging technique where frontier models teach procedural knowledge to smaller models via structured markdown files rather than weight compression, enabling cheaper inference while maintaining interpretability and versionability.
Thousands of specialized markets remain open for startups as major labs focus on prioritized categories, creating opportunity for AI harness companies to dominate vertical-specific use cases.
Observability, governance, and guardrails are critical for production agentic systems, requiring tracing, logging, evals as regression tests, and human-in-the-loop controls for high-stakes decisions.
Enterprise agentic systems require state persistence, checkpoints, and resilience mechanisms to handle crashes mid-workflow without losing progress across large-scale deployments.
Tool orchestration, validation, and safe exposure through registries (like MCP) will differentiate harness providers based on ability to safely expose tools and handle failures.
Context and memory retrieval systems will be bespoke per use case (radiology vs. paralegal), creating market opportunities for specialized retrieval infrastructure and context databases capturing business operating procedures.
AI is transitioning software from fixed-workflow SaaS to agentic systems requiring infrastructure for context retrieval, tool management, orchestration, and governance. This shift creates new competitive dynamics favoring specialized harness builders.
Security teams are transforming from people-focused to engineering-focused, with automation becoming essential to handle the scale of modern security challenges.
AI-assisted code generation and automated security processes will increase bug resolution velocity, making software more resilient despite increased code volume.
AI agents require fundamentally new security architectures focused on identity management and automated policy governance at scale, moving security from people management to engineering-driven systems.
Major platform companies like Microsoft and Databricks are competing to embed agent capabilities into their offerings to lock in customers and prevent workload migration to competitors.