Evan Steeg gave fantastic talk at Lean Culture on “Taking AI from Lab to Market: 35 years of closing the Gap.”
Blog From Lab to Market:
35 Years of Closing the AI Gap
For more than three decades, AI practitioners have faced the same fundamental challenge: bridging the gap between what works in the lab and what delivers reliable value in the real world. Evan Steeg draws on 35 years of experience spanning early neural network research, genomics AI startups, consulting, and enterprise transformation projects to explore why that gap persists and what organizations can do about it.
How to Make AI Pilots Successful
Evan offered advice on how to make AI pilots successful.
1. Speed Is Not Strategy
AI makes it easier than ever to build prototypes and MVPs, but building faster doesn’t guarantee business success. AI can tell you how to build something, but not necessarily what to build or why customers will buy it. Product-market fit still requires real customer conversations, testing, and validation.
2. Choose the Right Problem Before Choosing AI
Successful AI projects start with a clearly defined business problem, quality data, and measurable success criteria. Don’t implement AI simply because it’s exciting. Focus on problems where AI can deliver meaningful improvements in efficiency, revenue, risk reduction, or strategic capabilities.
3. Assess Organizational Readiness Before Launching
AI success depends on more than algorithms. Organizations must evaluate their data quality, infrastructure, employee skills, workflows, leadership, and governance. A technically successful AI pilot can still fail if the organization isn’t prepared to adopt it.
Download slides https://www.skmurphy.com/wp-content/uploads/2026/10/ESteeg-Lab_to_Market-v2.pdf
SKMurphy Take
Evan Steeg set a direction early in his career to leverage AI for biology, which is an extremely complex domain to build explanatory models for, much less predictive ones, and he persevered.
Evan offered an excellent high-level perspective on what it takes to move an AI application from a bench setup through testing to a full deployment. There were a number of good insights offered for what it takes to avoid common pitfalls on that journey.
Success factors: high-quality data, clear success metrics, high-quality people. Not all that different from most software projects except the data piece, which AI projects share with deploying machine learning models. It’s critical to manage data sources, monitor the data cleanup process, measure quality after cleanup, and monitor the impact of data updates. AI’s stochastic algorithms present a new challenge compare to traditional software and machine learning projects which rely on deterministic algorithms where the same inputs yield the same outputs.
Key failure modes
- Wrong Problem
- Data is not Clean
- No plan for adoption and scaling beyond lab
- Scaling too fast
- Measuring the wrong things, or not measuring at all
- Governance friction
Key aspects of well harnessed AI (Real-world AI agents fail to correctly follow instructions roughly 25% of the time in independent testing. Managing AI is primarily about paying attention.)
- Fixed tool registry
- Swappable models
- Deliberate context management
- Hard guardrails on tool calls and retries
- Logged agent loop
- Verification against trace not just model claims
Measuring AI Value
- Efficiency: Hours saved, error rates reduced, cycle times shortened, manual steps eliminated
- Revenue: New products or services enabled, faster time-to-market, improved conversion rates
- Risk Reduction: Compliance incidents avoided, fraud detected, decision quality improved, audit findings resolved
- Strategic Optionality: Capabilities built that enable future moves — platforms, datasets, and skills that compound over time
No value in creating a token furnace. These are standard metrics for most capability development or capability innovation projects. It’s still important to remember that an AI enabled capability must satisfy metrics in one or more of these areas. There is no inherent value in using AI per se.
How I got to know Evan: I curated this quote in September 2010″ “Organization and good planning are just crutches for those who can’t handle stress and caffeine,” that was credited to Evan Steeg. In August 2024 I went to use it again for a collection of quotes around “do it now because sometimes later never comes” and thought I should make sure he was a real person who acknowledged the quote. We connected on LinkedIn and I was impressed with his insights and invited him to speak at a Lean Culture.
Related Blog Posts
- Irina Calciu: Lessons from Building AI Systems, Products, Startups (Q&A)
- Bill Grosso on Startups after AI (webinar)
- Andrew Shindyapin: AI’s Impact on Software Development (Q&A)
- The AI Gold Rush: It Still Takes Teams (podcast)
- Mark Bennett on Using Claude Code for Application Development (Q&A)
- AI in Action: Practical Automation by Alex Panait (webinar)
- Matt Trifiro on Lessons Learned using AI for Marketing (webinar)
- Alex Panait on Current Trends and Possible Futures for AI (Q&A)
- Time to Market S01 E03 – How Will AI like ChatGPT Impact B2B Founders? (podcast)
Image Credit: “Complexity in the Real World” by Virpi Oinonen (@voinonen) Low resolution version used with attribution.