Software Abundance & Market Dynamics

• Software abundance — we will have 10x the amount of software available because of AI. TAM of Intelligence is Infinite makes the demand-side argument: cognitive work is input to almost every economic activity, so cheaper cognition creates new use cases rather than just replacing old ones. Inference Cost Collapse and Frontier Model Margin Expansion provides the supply-side mechanism: 90% YoY cost drops, but margins expanding because demand outruns price falls.
• Single use, ephemeral software and hyper-personalized software
• Software Has Made Us Soft
• There’s never been a better time to be a builder

Agents Replace Apps — New UX Paradigms

• Net-new UX/AI paradigms — dynamic UI, widgets and automations in the background that are personalized just to you. Claws - Persistent Looping Agents as App Replacement is the strongest example: Karpathy’s experiment collapsed six consumer apps into a single natural-language agent interface. The apps still exist underneath, but the user never touches them.
• System of Action - Evolution Beyond System of Record frames the architectural shift: software evolves from storing records to deciding and triggering actions. Whoever owns the system of action captures the strategic high ground.
• AXI Principles - Agent-Ergonomic Interface Design shows what tool design looks like when the agent — not the human — is the user. 10 principles benchmarked across 915 runs; interface design matters more than protocol choice.
• The Agent Is the Customer - A Convergence Thesis on Where AI Value Accrues ties it together: 5 independent sources converging on “value migrates from the record layer to the action layer.” The customer of software is increasingly an agent, not a human.

The Knowledge Work Jevons Question

• Knowledge work will undergo a Jevons Paradox — we will have more knowledge work, not less. But this is genuinely contested: Jevons Paradox vs Cognitive Displacement - The Unresolved Tension holds both the expansion argument (cheaper cognition → more demand) and the displacement argument (The Knowledge Work Cliff - Displacement of the Upper-Middle Class — the $80K-$400K class gets hit hardest).
• Most Work Is Scaffolding - The 75-99% Hypothesis sharpens the question: if 75-99% of knowledge work is tooling, formatting, and context assembly, AI automates the scaffolding and frees humans for the 1-25% of genuine cognitive work. Whether that expands or contracts total employment depends on which side of the Jevons tension wins.
• Abundant vs Scarce After AI - The Bifurcation of Post-Scarcity maps what becomes cheap vs what stays valuable.

Switching Costs Decline & Pricing Shifts

• Switching costs will decline as a moat; way easier to automate pipelines. Seat-Based to Token-Based SaaS Pricing Transition describes the mechanism: customer AI savings show up as fewer seats first, and new budget flows to tokens, consumption, outcomes.
• Grow 10 or Earn 40 - Two Paths for Mature Software Companies is the consequence: the middle disappears. Either accelerate growth by 10+ points or rebuild to 40%+ true operating margins.
• The counter-argument: Distribution as the Remaining Moat - Why SaaS Incumbents Aren’t Dead — code can be cloned cheaply but distribution cannot. Healthy incumbents hold; complacent ones get flipped. The diagnostic is engineering velocity, not framework allegiance.

Everyone Becomes an Engineering Manager

• There will be way more software engineers, as cost of building goes down
• Everyone will be an engineering manager, with AI engineering agents. Token Throughput as the New Coding Bottleneck makes this concrete: when agents do the typing, the bottleneck shifts from cognitive bandwidth to parallel session orchestration.
• Research Org as Tunable program dot md takes it further: the org itself becomes a markdown document you can version, A/B test, and meta-optimize. Engineering management becomes configuration, not supervision.
• Expertise Diffusion - The One-Way Ratchet — expert knowledge gets systematically extracted into documentation, SOPs, and training data. The bar to entry drops; the moat is learning rate, not knowledge stock.
• Engineering will adjust and evolve, like from assembly to Python
• Bar for product quality will rise

How It Plays Out — The Deployment Pattern

Absorb Automate Unbundle - Three Phases of Technology Deployment provides the temporal frame: first organizations absorb AI into existing workflows, then automate the obvious, then — the phase that actually reshapes industries — unbundle things that were only bundled because of constraints AI removes.

The vertical software version: Run the Business vs Do the Work - The AI-Era Vertical SaaS Shift — historical SaaS helped owners run the business; AI-era SaaS helps practitioners do the work. Hero User Strategy - Native AI’s Wedge Into Vertical Software is how native AI breaks in. Full Stack Down vs Full Stack Up - Two Directions for AI Application Vertical Integration maps the two integration directions. Neofirms - AI-Native Professional Services as a New Category is the new category this creates.

• LLMs aren’t the end-all answer to core intelligence, that’s still an unsolved research problem
• Cognitive Automation Accelerates the Robotics Timeline — the atoms-vs-bits timeline may be shorter than expected because cognitive labor in robotics R&D is itself being automated

Related Notes

• AI and Investing Thesis — investment implications of these hypotheses
• AI + Business Model — how business models evolve
• Agent Harnesses — the orchestration layer where most value lives
• Context Engineering — the engineering discipline for the agent era

References

• Future of software X thread
• Levie on AI
• Salesforce Loses Its Head — Saanya Ojha — Salesforce Headless 360 as case study of a defining SaaS company restructuring for agent-operated access