Guest Feature: Sandeep Krishnamurthy, Dean, Cal Poly Pomona and Robert Barrios, CIO, GALLO
The Economic Value of an Idea
We built the knowledge economy with the idea as the fundamental economic unit. The creative economy meant that those with the best ideas got funded and then launched.
Then, AI collapsed the cost of the first draft. Compare to a world where “the back of the envelope” represented hours of toil leading to an initial concept. In today’s world, AI can generate ten new ideas based on some rough guidance from a creator. There is work in choosing among these competing ideas and coming up with a clear concept.
Once the concept is finalized, a developer paired with business expertise can move from concept to working prototype in about an hour. That sentence still feels strange to write because two years ago it would have been absurd.
The Numbers Tell the Tale
The quantitative insights that we are seeing already match the anecdotal reality now. A controlled GitHub Copilot study found developers using AI completed coding tasks 55.8% faster, McKinsey estimated 20-45% productivity improvements with some tasks completing up to twice as fast, and more recent applied research shows time savings of 30-50% across documentation, debugging, unit tests, and pair programming.
This isn't just a developer productivity story, it's a business formation story. Y Combinator recently claimed that 25% of startups in the most recent batch had codebases that were 95% AI-generated- the clearest signals that AI had moved from code assistant to company-building accelerator. The exact meaning of "95% AI-generated" deserves scrutiny, but the strategic implication is hard to miss because the cost of getting to a working product artifact has fallen dramatically.
The unit economics of generating the MVP just blew up.
The Real Work is in Going to Production-Ready
While the production velocity is real, there is a big gap between "working demo" and "production-ready." This is the defining constraint.
As of this writing, the ratio from MVP to production-ready deployment is averaging about 1:2 across the teams I work with. One hour to build the working prototype, then two more hours to get it into a state where the person who owns the application says it's supportable, secure, and performant enough for real users at scale. That ratio holds when you have mature automated pipelines, established deployment processes, and teams that already know how to operate production systems.
Without those foundations, the ratio gets worse. Much worse.
The scale of what's happening upstream makes this urgent. GitHub's COO Kyle Daigle shared that there were 1 billion commits in 2025, and in 2026 the platform is seeing 275 million commits per week, which puts it on pace for 14 billion this year. GitHub Actions went from 500 million minutes per week in 2023 to 1 billion minutes per week in 2025. The production engine is running hotter than ever, and the question is whether your pipes can handle the throughput.
The best analogy is the difference between building a prototype car and building the assembly line. The prototype proves the idea can work, it creates excitement, and it lets people touch the concept. But the assembly line proves that the car can be produced repeatedly, at quality, under cost constraints, with safety standards and process discipline. AI has made prototype cars dramatically easier to build, but it hasn't magically built the assembly line.
Think Digital Infrastructure - not islands of vibe code
This is why the phrase "vibe coding hangover" has resonated with technology leaders. Fast Company reported in September 2025 that while vibe coding is useful for quickly assembling demos, AI-generated code becomes difficult to maintain, debug, secure, and explain once it enters real production environments. One senior engineer described AI-generated code as potential "development hell" when it creates technical debt that humans must eventually pay down.
The research backs this up. One 2025 study found that after Copilot adoption, experienced core developers reviewed 6.5% more code while their own original code productivity dropped by 19%, which suggests that AI-generated output can shift the burden from creation to review, repair, and maintenance. AI may make more people productive at the edge of development, but if the resulting code increases the load on senior engineers, you haven't solved a productivity problem. You've moved the bottleneck.
Absorptive Capacity of Innovation
The central question is no longer "can we build it?" The better question is "can we absorb it?" Can the organization take AI-generated output and move it through security review, testing, integration, documentation, deployment, monitoring, support, and continuous improvement without creating chaos?
By "pipes" I mean the unglamorous infrastructure that determines whether AI output turns into enterprise value. Automated testing, CI/CD pipelines, security scanning, API management, data governance, cloud architecture, rollback procedures, observability, documentation standards, human code review, change management, incident response, and compliance workflows. These don't look magical in an AI demo, they rarely produce applause, but they increasingly determine whether an organization can benefit from AI at scale.
In the pre-AI world, slow development cycles concealed weak organizational systems because when everything moved slowly, teams had time to compensate manually with meetings, spreadsheets, heroics, and institutional memory. AI compresses the cycle and removes the cushion. When software creation accelerates, every downstream weakness becomes visible. If your team can generate ten promising product ideas in a week but can only safely deploy one in a quarter, your bottleneck isn't creativity. If developers produce code faster than security can review it, the problem isn't developer productivity. If business units imagine AI-enabled workflows but your data architecture can't support them, your constraint isn't ambition, it's infrastructure.
Those with the better pipes win
What's interesting is that the organizations absorbing AI output fastest right now aren't the ones with the biggest AI budgets. They're often the ones that invested years ago in DevOps, cloud modernization, automated testing, modular architecture, data governance, and change management maturity. That work seemed boring at the time, and now it looks prophetic because plumbing compounds. The organizations that treated infrastructure as strategic are turning AI velocity into business velocity, and the organizations that treated infrastructure as back-office overhead are discovering that AI produces more unfinished work than their systems can absorb.
For leaders, the question must shift. The old question was "how fast can we build?" and the new question is "how fast can we responsibly deploy?" That word "responsibly" matters because it includes security, privacy, accessibility, reliability, compliance, maintainability, observability, and customer trust.
A working demo isn't a product. A product is something that survives real users, real data, real edge cases, real attackers, real audits, real customer expectations, and real operating pressure. When you see a dazzling AI-generated prototype, the wrong response is "why can't this be live next week?" and the better response is "what would it take to make this production-grade?" That one question changes the conversation from theater to capability.
Finish Strong
AI has dramatically lowered the cost of starting, which is extraordinary because more people can build, more ideas can be tested, more prototypes can be shown to customers, and more business analysts and product managers can participate in software creation. But AI hasn't automatically lowered the cost of finishing because finishing still requires engineering judgment, testing, security, architecture, maintainability, and human accountability.
Starting is easier than ever, and finishing is now the differentiator.
If your teams are producing faster than your infrastructure can deploy, that's your signal. The answer isn't more AI tools, it's better pipes. The next phase of AI competition won't be won by whoever has the most prototypes, it'll be won by whoever can turn prototypes into production, production into adoption, and adoption into measurable value.
That requires deployment infrastructure, automated testing, security-by-design, modern data architecture, DevOps maturity, and change management systems that can move at the speed of AI without sacrificing trust.
The MVP is easy. The pipes are the strategy.
