Building software has never been cheaper. Ninety-two percent of U.S. developers now use AI coding tools daily, and 46% of all new code is AI-generated. What that abundance has not touched is distribution; the cost of getting a buyer to trust, integrate, and standardize on your product, which has always been the harder problem in enterprise software.

Aaron Levie, co-founder and CEO of Box, described it on X : "The plurality of costs in most enterprise software companies is actually on GTM, because at scale most enterprise software categories are tough to break into and need a heavy amount of consultative selling and support for implementation and integration of solutions." AI, he argued, has not reduced that need. In many cases it has increased it.

The venture capital market has bet heavily on the build side. Over $1 billion flowed into AI coding platforms in 2025 alone, and Replit raised $400 million at a $9 billion valuation in March 2026, tripling its valuation in six months. That capital reflects a genuine insight: development velocity has increased. The question investors are slower to answer is what happens to the rest of the unit economics when code becomes a commodity.

The answer is visible in SaaS benchmarks. GTM expenditures run 40% or more of revenue at high-growth cloud software companies, according to a BCG survey of more than 90 executives at mid-sized B2B software firms. Enterprise sales cycles average six months or longer , and customer acquisition costs for enterprise deals run five to ten times higher than the B2B SaaS median. Cheaper code generation does not move any of those numbers.

Levie framed the dynamic as a conservation law: "If you make one thing cheaper and more abundant (development of software) then the new problem of discoverability and market differentiation (GTM) becomes the hardest part." The supply side has never been more liquid. App Store submissions jumped 84% year over year in Q1 2026, to 235,800 new apps in a single quarter. Apple expanded review times from 24-48 hours to 7-30 days to cope. More software means more noise for every enterprise buyer navigating a decision.

That buyer problem compounds in regulated industries. Levie noted in a March 2026 roundup of conversations with more than 20 enterprise AI and IT leaders that implementation timelines have not compressed with AI capability gains. Enterprises still require security reviews, integration with existing data infrastructure, legal sign-off, and change management before any software gets into a production workflow. None of that has an AI shortcut.

The crowding at the application layer also changes what signals buyers can use for differentiation. When 60% of new code is AI-generated and any founder can ship a working prototype in a weekend, product features are no longer a reliable filter. Trust, distribution, reference customers, and the depth of implementation support become the differentiating variables; each of which requires time and relationships, not compute.

Karri Carlson, a product executive who replied to Levie’s post, flagged a downstream risk that VCs have not fully priced: what happens when AI agents make software procurement decisions? Machines, she argued , will only value trust and taste if those qualities can be quantified and compared. That scenario is not imminent, but the category of founders building for an agentic procurement future, where discoverability requires machine-readable signals rather than relationship-driven ones, has an edge today.

For investors allocating to the application layer, the implication is that GTM infrastructure is underpriced relative to dev tooling. Companies that have cracked low-cost distribution; through PLG (product-led growth), embedded integrations, or community-driven adoption, carry durable advantages that AI cannot replicate at the same rate it replicates features. The moat in enterprise software was never the code. It was always the trust required to run on someone else's infrastructure, and that cost structure is, if anything, going up.