Key Takeaways
- • Andreessen Horowitz's report "How AI Applications Earn Their Premium," published November 2, 2026, argues AI applications sustain premium prices only when they deliver more value than customers can build in-house.
- • The report organizes its case around three principles: outperform the customer across the full workflow, make reliable context cheaper to obtain than to rebuild, and demonstrate completed work's value before pricing it.
- • Repsol's legal team piloted Harvey over eight weeks with 50 lawyers across 12 jurisdictions, finding it superior to Microsoft Copilot and Repsol's internal prototype Lexia on accuracy and time savings, per the report.
- • Harvey reports weekly savings of four to six hours per lawyer and 96% adoption at Repsol following the department-wide rollout.
- • In an August 2026 survey of 100 enterprise LLM decision-makers, 94% expected total LLM spending to rise and 37% expected it to at least double over the following year.
Summary
AI applications can sustain premium prices when buying them delivers more value than customers can achieve by building their own alternatives, according to Andreessen Horowitz's "How AI Applications Earn Their Premium" report, published November 2, 2026. As models improve and development becomes easier, vendors face a tougher test of what customers will keep paying for. The report identifies domain expertise, deep integrations, dependable execution and hard-to-access context as the foundations of that value, while noting that stronger models and better development tools make individual features easier to replace.
The report argues that a specialist vendor must compete on the complete workflow rather than on model access alone. A customer need not replace an entire application to pressure its price; bringing a frequently performed task in-house may suffice. Repsol's legal team tested this proposition: a blind comparison ranked Harvey ahead of Microsoft Copilot and Repsol's internal prototype Lexia on output quality, and an eight-week pilot involving 50 lawyers across 12 jurisdictions found Harvey superior on accuracy and time savings. Repsol subsequently rolled Harvey out across the department, and Harvey reports weekly savings of four to six hours per lawyer and 96% adoption.
On context, the report distinguishes the value of exclusive access from the value of preparation. Where customers already have access to information, the advantage lies in the effort needed to keep it accurate, current and usable. One industry-specific software company interviewed for the report has accumulated proprietary pricing benchmarks covering roughly 3,000 companies and is exploring a Model Context Protocol (MCP) server to expose them to AI systems. The report says eliminating duplicated work can lower customers' total costs while producing a margin for the vendor.
The report also calls for demonstrating value through units of completed work, such as a reviewed contract, resolved support conversation or enriched record, with delivery costs including retrieval, model calls, tools, retries, evaluation and human review. It cites Decagon's case with Notion, which chose an external provider over building internally and reports up to a 34% improvement in ticket resolution time. Clay's pricing memo, as cited in the report, separated Data Credits from orchestration Actions, kept fixed credit prices for bounded AI workloads and passed through token costs without markup for variable reasoning models.
The report concludes that the final test is whether an application offers a better combination of total cost, quality, speed and risk than the customer's strongest credible alternative. As improving models strengthen those alternatives, the report argues that sustaining a premium requires the application's advantage to keep expanding.
The report argues that a specialist vendor must compete on the complete workflow rather than on model access alone. A customer need not replace an entire application to pressure its price; bringing a frequently performed task in-house may suffice. Repsol's legal team tested this proposition: a blind comparison ranked Harvey ahead of Microsoft Copilot and Repsol's internal prototype Lexia on output quality, and an eight-week pilot involving 50 lawyers across 12 jurisdictions found Harvey superior on accuracy and time savings. Repsol subsequently rolled Harvey out across the department, and Harvey reports weekly savings of four to six hours per lawyer and 96% adoption.
On context, the report distinguishes the value of exclusive access from the value of preparation. Where customers already have access to information, the advantage lies in the effort needed to keep it accurate, current and usable. One industry-specific software company interviewed for the report has accumulated proprietary pricing benchmarks covering roughly 3,000 companies and is exploring a Model Context Protocol (MCP) server to expose them to AI systems. The report says eliminating duplicated work can lower customers' total costs while producing a margin for the vendor.
The report also calls for demonstrating value through units of completed work, such as a reviewed contract, resolved support conversation or enriched record, with delivery costs including retrieval, model calls, tools, retries, evaluation and human review. It cites Decagon's case with Notion, which chose an external provider over building internally and reports up to a 34% improvement in ticket resolution time. Clay's pricing memo, as cited in the report, separated Data Credits from orchestration Actions, kept fixed credit prices for bounded AI workloads and passed through token costs without markup for variable reasoning models.
The report concludes that the final test is whether an application offers a better combination of total cost, quality, speed and risk than the customer's strongest credible alternative. As improving models strengthen those alternatives, the report argues that sustaining a premium requires the application's advantage to keep expanding.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.