Tingyu Su on why AI startups should treat the founding designer as a strategic hire from the start of company building

Stanford research says AI performance on a key coding benchmark rose from ~60% to nearly 100% in one year, while organizational AI adoption reached 88%. The article frames this as rapidly improving ability to turn ideas into working software products, with the main challenge shifting from capability limits to broader execution.

Analysis

The near-term market winner is not “software” broadly but the stack that monetizes developer throughput: hyperscale cloud, AI tooling, and endpoint/security vendors that sit on top of a larger code footprint. Faster shipping usually means more deployed instances, more model calls, and a wider attack surface, so adoption can translate into higher consumption even if headcount growth slows. That makes the earnings asymmetry better for MSFT/AMZN/GOOGL than for most application-layer SaaS names where feature velocity is the main moat.

The clearest losers are labor-arbitrage IT services and outsourced engineering firms such as EPAM, CTSH, INFY, and WIT. Their pricing model depends on billable hours; AI compresses hours per unit of output before it raises demand enough to offset the hit, so utilization pressure can show up first in margins, then in hiring, then in guidance. Over 1-3 months, watch for commentary on project delays and pilot-to-production conversion; over 6-18 months, the risk is structural share loss as customers internalize more dev work.

The consensus is probably too focused on productivity gains and not enough on supply expansion. If building software becomes materially cheaper, the constraint shifts from engineering capacity to distribution, differentiation, and trust, which should favor the biggest platforms and punish undisciplined SaaS with weak retention. What would falsify the bearish services view is evidence that AI lifts deal size and utilization simultaneously; what would falsify the bullish platform view is cloud consumption decelerating despite higher code output, implying the efficiency gain is real but monetization is not.

The biggest second-order effect is that cheaper code likely creates more software, not fewer engineers, but it also creates more commoditized software and more churn. That argues for a barbell: long infrastructure toll collectors, short labor-intensive implementers, and selective caution on mid-cap software where multiple compression can arrive before any productivity benefit shows up in margins.

Key Decisions for Investors

• Over the next 1-3 months, buy MSFT or AMZN on any post-earnings weakness as a proxy for rising AI-driven developer throughput and higher cloud/inference consumption; risk/reward improves if management commentary confirms higher internal and customer code generation.
• Pair trade: long MSFT / short EPAM or CTSH over 3-6 months. Thesis: AI compresses billable-hours economics faster than it lifts outsourcing demand; stop if services firms reiterate sustained utilization expansion or pricing power.
• Watch list: short INFY or WIT into any rally if the next two quarterly prints show slower headcount growth than revenue growth. The market may be underpricing margin compression from AI substitution in offshore development.
• Selective long on security/observability names such as CRWD or DDOG if valuations reset; more code and more deployments should raise attach rates, with the catalyst visible in 1-2 quarters of consumption data.
• No immediate options trade on generic software beta unless the next earnings season confirms lower hiring. If you want optionality, use a small call spread on MSFT versus a basket short of IT services as a catalyst-driven relative-value expression.