Conviction
Living thesis
A working set of convictions on where software design is headed, how teams should operate with AI, and what changes when the IDE stops being the center of gravity. Each section links to essays and notes that argue the point — conviction to evidence in one click.
Table of contents
The future of software design
Design is shifting from artifacts to outcomes. The work moves upstream: defining systems, interfaces, and constraints that enable software to be assembled, not just designed. AI accelerates this shift by compressing the distance between intent and implementation.
- Systems over screens: Designers specify behaviors, contracts, and data models that span products and platforms.
- Proof over prototype: Real, runnable proofs replace static mockups; fidelity is functionality.
- Human-in-the-loop: Designers become editors and directors of AI-generated variants, curating toward taste and performance.
Further reading
- A new MVC is emergingHow models, views, and agents recompose the design surface
- Building your designsProof over prototype — shipping as the design medium
- Dynamic InterfacesGarden notes on interfaces that adapt in real time
AI-native management
Teams are evolving from maker-only to maker–model ensembles. Management shifts from headcount planning to capability planning—what combination of people, data, and models delivers outcomes reliably and safely.
- Orchestration: Managers design workflows where agents handle repetitive work and people handle ambiguity.
- Quality as a product: Evaluation data, benchmarks, and red-teaming become first-class assets.
- Ethics and controls: Guardrails, consent, and attribution are operating requirements, not post-facto checks.
Further reading
- Operator ModeDoing the work between founder and professional management
- Career rebootReorienting leadership craft for the AI incursion
- Design's Seat at the (Cap) Table: Part IWhy design judgment belongs in strategy and ownership
Post-IDE world
As coding shifts from manual keystrokes to conversational and agentic workflows, the IDE becomes a collaboration surface rather than the primary tool. The primitives are problems, tests, and contracts; the output is orchestrated by agents and reviewed by humans.
- Intent-first: Problem decomposition, specs, and evaluation drive generation.
- Continuous verification: Tests, types, and instrumentation are the new UI.
- Composable agents: Tool-using agents coordinate across repos, clouds, and runtimes.
Further reading
- IDDE: The infusion of design and dev toolsKilling handoff in favor of shared creation surfaces
- The formlessness of AI agentsFinding vessels for capabilities that outgrow the IDE
- A New MVC is EmergingEarly notes on LLMs, apps, and agents as the new stack
Human × AI collaboration
Great outcomes emerge when people and models play to their strengths. Humans set direction, define taste, and handle ambiguity; AI accelerates exploration, handles repetition, and scales evaluation.
- Taste as direction: Humans define the bar; models explore the possibility space toward it.
- Interfaces as partnership: The UI is no longer a fixed shell — it mediates intent between people and agents.
- Creative leverage: AI expands range without replacing judgment, craft, or accountability.
Further reading
- Waiting for AI's pull-to-refresh momentThe race to invent the interaction grammar of AI interfaces
- AI as Creative PartnerEarly thoughts on collaborating with generative tools
- Real-time strategy and AI interfacesRTS mental models for directing agents in the field
Keep exploring
- WritingLonger essays across craft, career, and AI
- NotesThe digital garden — sketches that feed the thesis
- Proof of ConceptWeekly writing where many of these ideas first shipped