Tomasz Tunguz

Hybrid systems using deterministic code for predictable tasks and LLMs only for ambiguous reasoning outperform fully agentic approaches, with 65% of workflow nodes running as non-AI code after optimization.

“Author of "Is AI Doing Less & Less?" in Tom Tunguz”

Practitioners building AI agent systems should prototype with state-of-the-art models, fine-tune specialized smaller models for well-defined tasks, use statically-typed languages to reduce hallucinations, and implement closed-loop prompt optimization systems for incremental performance improvements.

“Author of "9 Observations from Building with AI Agents" in Tom Tunguz”

Google achieved a 52x year-over-year increase in AI token processing to 10 billion tokens per minute while reducing serving costs by 78%, demonstrating unprecedented efficiency gains in its Gemini platform.

“Author of "Google's 52x AI Growth"”

Microsoft has built a $281B revenue contract with OpenAI, representing 45% of its commercial backlog, demonstrating massive AI demand but creating significant customer concentration risk for even the largest technology companies.

“Author of "$281B From One Customer" in Tom Tunguz”

Gemini 3's breakthrough performance improvements with identical parameter count to Gemini 2.5, combined with Nvidia's Blackwell infrastructure dominance, proves scaling laws remain viable and the 2025 'scaling wall' narrative was unfounded.

“Author of "The Scaling Wall Was A Mirage" in Tom Tunguz”

OpenAI's $1.15 trillion infrastructure commitment between 2025-2035 implies the company must grow revenue from $10B in 2024 to $577B by 2029 to maintain target profit margins, requiring a scaling trajectory comparable to Google's historical growth.

“Author of "OpenAI's $1 Trillion Infrastructure Spend"”

AI tool design must prioritize complex, parameter-rich tools over simple ones to improve LLM accuracy and token efficiency by up to 70%, fundamentally shifting from human-intuitive to AI-cognition-optimized architecture.

“Author of "How AI Tools Differ From Human Tools"”

AI vendor growth is creating second-order effects throughout the software stack, with MongoDB's Atlas experiencing renewed growth acceleration as AI-native companies adopt it for vector search and data management at scale.

“Author of "The Second-Order Effects of AI"”

Perplexity AI made a surprise unsolicited $34.5 billion bid to acquire Google Chrome, which industry investors dismissed as a stunt worth far less than the browser's true value.

“Venture capitalist who valued Chrome at ten times the Perplexity bid or more.”

“Tomasz Tunguz wrote analysis on how Anthropic and OpenAI have boosted revenue through strategic customer segmentation and pricing changes.”

“Byline author discussing unbundling of AI capabilities and specialized models for production.”

“Author of 'What Does Pacing Mean?' essay examining five different interpretations of AI pacing proposals by industry and government actors.”

“Venture capitalist who wrote analysis of financing requirements for AI infrastructure expansion requiring $5 trillion in debt.”

“Wrote about the frontier AI market sorting into closed camps with rationing programs and whitelists.”

“Byline author analyzing Nvidia's record quarterly revenue and growth trajectory.”

“Wrote article analyzing the Bullwhip Effect in AI infrastructure supply chains as demand shocks cascade upstream.”

“Tom Tunguz tested local models like Qwen3.8-27B against cloud models and found that smaller local models can outperform larger cloud models through reasoning.”

“Author of newsletter article explaining how test-time training allows AI models to adapt by updating weights during use, reducing memory needs but increasing computational demands.”

“VC founder commenting that frontier models aren't required for most enterprise demand.”

“Led Ollama's Series B; stated that platform layer for open models is one of most valuable positions in software.”

“Published a working list of companies that have switched off frontier closed models in production.”

“Author of blog post analyzing Microsoft's new metric of average token usage on model release cards and its implications for AI pricing.”

“Author analyzing AI pricing strategies of Google, OpenAI, and Anthropic and their relation to cash availability.”

“VC and writer who conducted a benchmark testing local 35B models versus Claude Opus on eight agentic tasks, finding local models can handle half of daily work tasks.”

“Author analyzing how the AI inference market is fragmenting into latency tiers, multimodal models, and edge models, creating room for multiple winners.”

“Byline author of Tom Tunguz newsletter discussing scarcity in AI compute and reshaping of access to cutting-edge models.”

“Authored analysis comparing Anthropic's growth trajectory to NVIDIA's market cap and other software companies.”

“Byline author analyzing AI companies' shift from unbundled point solutions to bundled platforms driven by rapid model changes.”