Notes on AI Apps in 2026
TLDR
- A16z’s Anish Acharya argues cheap code hasn’t yet diffused across the enterprise, and less than 10% of its implications for company building have been realized.
Key Takeaways
- All current knowledge-work tools (IDEs, Figma, spreadsheets) target execution, not exploration; Acharya expects a new category of “thinking tools” to emerge as coding agents handle more execution.
- Cursor is named as the furthest along in shifting toward exploration-first design; Antigravity is cited as “agent first” in product approach.
- Every enterprise function, including legal, finance, and HR, should become software-first; domain-specific products like Harvey and bare-metal agents like Codex and Claude Code are named as the two paths.
- AI-native apps are diverging from models by combining multi-model orchestration, domain-specific UI, and broad feature surface; coding app startups generated over $1 billion in new revenue in 2025.
- Labs and big tech face prioritization constraints (OpenAI competing across consumer, enterprise, model, and hardware simultaneously; Google’s regulatory commitments) that create durable space for app-layer startups.
Why It Matters
- The culture change required for enterprises to become software-first in every function is described as equally hard as the organizational change, not a solved problem.
- Consumer access to AI capabilities remains limited by command-line UIs; Wabi and ChatGPT/Grok image tabs are cited as early catalysts for broader consumer adoption.
- If coding agents handle execution, the bottleneck shifts to deciding what to build; current models are characterized as producing bland, derivative product ideas, making “thinking tools” the open problem.
Anish Acharya, Andreessen Horowitz · 2026-01-08 · Read the original
