Insights
Six theses, each starting as a concise claim and expanding into a visual model and a dedicated page. They are the reasoning behind our roadmap; the home page is the short version.
AI software development is evolving through levels of control, from manual coding to deterministic toolchains and formal domain systems. The durable advantage comes from governed leverage, not autocomplete alone.
Read the full thesisSmarter models still help, but today’s software lifecycle imposes a practical ceiling. The opportunity shifts toward contextual specialization, lower token dependence, and cost-efficient execution on infrastructure teams already own.
Read the full thesisAgentic coding adoption is splitting between pipeline-native systems outside the developer PC, IDE-native agentic workflows inside it, and mainstream augmentation built on commodity models. Crossing both chasms requires repeatable, governed systems, not assistant usage alone.
Read the full thesisSoftware engineering is harder than coding because organizations rarely share an operational definition of work well done across security, privacy, resilience, compliance, cost, and value. AI's deeper opportunity is to make those trade-offs more consistent, transparent, and auditable.
Read the full thesisAI software is increasingly constrained by the economics of tokens, total cost of ownership, and infrastructure scarcity. The durable advantage shifts toward forecasting consumption of tokens with Moltke, measuring detailed usage with Solon, reducing opacity, and choosing architectures companies can actually afford and secure capacity for.
Read the full thesisEnterprise AI value is shifting away from raw model power toward orchestration, memory, tools, security, agentic workflows, and architectures tailored to real operating constraints. The durable product advantage lives in the governed system around the model.
Read the full thesis