Tomasz Tunguz

Tomasz Tunguz

General Partner at Theory Ventures; AI impact on SaaS economics & venture metrics.

Voice: Data-led SaaS analyst voice. Each post anchored to a specific benchmark chart or financial ratio (Rule of 40, magic number, sales efficiency, payback period). Concise declarative titles ('X is Y'). Concrete numbers from portfolio observations or public 10-Ks. Reuses named frameworks across posts. Short paragraphs, often a question prompt at the end.

Frameworks

• AI agents require seven disciplines of domestication to move from powerful but wild capabilities to production systems: context & memory, tools & action, orchestration & loop, state & persistence, sandbox & compute, observability & governance, and cost & workflow optimization.
• Modern AI systems achieve better outcomes through architectural separation: code handles predictable, routine operations while AI focuses exclusively on complex, judgment-intensive tasks like summarization and decision-making.
• Software companies operate under a structural dichotomy: high gross margins offset by heavy sales/marketing/R&D spending (growth model), versus AWS's infrastructure-led capital efficiency model (profitability model). Profit margins and revenue growth show no correlation in the software model.
• AI value measurement is shifting from single-dimension performance benchmarks to dual-axis assessment (performance AND cost per token), which cascades up the stack: models compete on intelligence per dollar, applications compete on dollars per outcome.
• Skill distillation: frontier models author procedural markdown files that smaller local models execute, transferring institutional knowledge through inspectable procedures rather than compressed weights. This creates a three-layer architecture (knowledge base → skills → agent loop) where the teacher-student relationship is mediated by versionable instructions.
• Agent gravity describes how platforms compete to retain AI agent workloads and associated data processing, as agents create stickiness by deciding where to run and process data. Platforms that capture more agents and their data flows build compounding gravitational advantage.
• Software systems are evolving from single fixed interfaces to dynamic multi-interface systems ('many heads'), where AI generates contextually appropriate UIs (audio, web app, spreadsheet) on demand, with the core value shifting to interface control/validation and artifact/context management over time.

Principles

• As AI accelerates both attack and defense velocity, security teams must transition from people management to policy engineering, with every agent requiring identity-based governance at the point of action—creating a narrowing window of exploitability as automated patching outpaces automated attacks.
• Companies that enable pre-IPO secondary liquidity reduce selling pressure at the public offering, changing the traditional IPO dynamics exemplified by older comparisons like Alibaba and SoftBank.
• AI-driven software revenue disruption creates cascading risk through debt financing structures, making leveraged software companies particularly vulnerable to earnings decline.
• AI's impact on software sectors follows a demand-side pattern: sectors where AI increases customer demand (data infrastructure, security) outperform those where AI makes customers more efficient (vertical software, workflow tools).
• When tech companies grow revenue while reducing headcount, revenue per employee becomes a primary efficiency metric, with AI-enabled productivity fundamentally resetting baseline expectations across the industry.
• Technological productivity breakthroughs that speed up production processes don't eliminate jobs but instead expand employment by creating new adjacent markets and roles across the value chain.
• Using AI feedback loops to systematically improve interpersonal communication creates measurable behavior change in meeting effectiveness.
• Short interest concentration reveals market skepticism's true target: when shorts concentrate in capital-dependent AI plays rather than uniformly across the stack, it signals specific doubts about business model viability rather than broad AI fatigue.
• Pricing changes in AI models signal strategic shifts: price cuts when cash is plentiful and market share matters, price increases when cash is tight and margins matter.
• Skills-based abstraction layers increase software accessibility by reducing learning complexity, but security verification becomes the critical gating factor for adoption.

Predictions

• When large private companies with low float IPO into major indexes, forced index buying triggers significant market reallocation effects. Three huge companies—SpaceX, OpenAI, and Anthropic—plan IPOs worth nearly $3 trillion combined with very few shares to the public, causing major market shifts.

Voice samples

• "The end of the software era is the beginning of the harness era. Like a mustang, AI is powerful but wild."

• "Benchmarks are now measured on two dimensions: overall performance & the cost to achieve that intelligence."

• "If data gravity was the most important force in the Decade of Data, agent gravity will be the same in the Decade of Agents."

• "AI is just as powerful for defenders as it is for attackers. The fear narrative underestimates this fact."

• "Models compete on intelligence per dollar, applications compete on dollars per outcome."

• "My personal agent runs my inbox, deal pipeline, blog publishing, calendar, & research. It looks less like a chatbot & more like a small operating system."

Generated from 26 items, 24 kept after dedup. Full attribution: logs/tomasz-tunguz.jsonl.