In this episode of the Disambiguation podcast, host Michael Fauscette talks with Chris Fitkin, Co-Founder and Partner at Metacto, about why most companies are spreading AI experiments too thin across every department, why that wide-and-shallow approach produces shelfware instead of results, and what it actually takes to get AI into production in the mid-market.Chris has 25 years in software engineering, a master's in software engineering, and is an AWS Certified Solutions Architect. He has been a CTO in cybersecurity, led due diligence at private equity firms, and co-founded Metacto with Garrett Fritz. The firm started in mobile app development and fractional CTO work, then pivoted to operational AI for mid-market and private equity-backed companies after seeing how dramatically the landscape was shifting.The conversation covers the AI sprawl problem (88% of companies are using AI in at least one function per McKinsey, but only 5% see measurable impact per MIT), the repeating cycle of cool demo to mixed results to low adoption to shelfware, why companies that go narrow and deep are more than twice as likely to see measurable bottom-line impact compared to those that scatter experiments across the organization, the Five Signals framework for picking your first AI workflow (email, spreadsheets, copy-paste relays, contactless approvals, repeat expert answers, report factories), why mundane workflows are the right starting point, the 5%/95% gap between demo and production (access control, business rules, quality checks, human review, audit trails, monitoring, versioning, ownership), failure modes that compound when nobody catches a bad LLM decision for five days, context engineering and the three parts of building good context (transactional data, document corpus, codified business rules and domain knowledge), why business first has to replace AI first, the mid-market pricing reality (Anthropic offered one client $1.2 million a year in token minimums, Mars Inc pays $600,000 a month to Google Gemini), the AI Engineering Maturity Index assessment tied to EBITDA and enterprise value, and practical advice for stuck leaders.Timestamps:02:07 - Change is the only constant across 25 years of technology cycles03:19 - App Store submissions doubled while downloads decreased04:16 - AI sprawl: 88% using AI, only 5% see measurable impact05:09 - The shelfware cycle: cool demo, mixed results, low adoption07:20 - Shadow AI is the new shadow IT07:43 - Narrow and deep is 2x more likely to produce bottom-line impact08:43 - Going deep: problem first, define success metrics before you build11:14 - Solution looking for a problem versus problem looking for a solution12:27 - Five Signals framework for picking your first AI workflow13:03 - Mundane workflows are validated by human capital investment14:31 - The 5%/95% gap: demo is 5% of the work, production is 95%16:05 - Failure modes: bad decisions compounding, admin database access exposed16:59 - Context engineering versus prompt engineering19:06 - Transactional data: clean, current, deduplicated data warehouse19:27 - Document corpus: proposals, QBRs, deliverables tagged with recency20:04 - Business rules and domain knowledge: codifying what lives in people's heads21:32 - Business first, not AI first: product managers lead engagements22:25 - Requirements engineering: the discipline everyone is rediscovering24:01 - The mid-market gap: too small for McKinsey, too complex for license distribution26:42 - Start small, measure lift, use wins to fund the next projects27:51 - AI Engineering Maturity Index: 30-day assessment tied to financial metrics33:54 - Focus on people: dedicated time, builders and advocates, adoption training36:52 - Leading a hybrid workforce: managing human and digital workers38:33 - Recommendation: Reid Hoffman's Masters of Scale with IBM CEO Arvind KrishnaGuest: Chris Fitkin, Co-Founder & Partner, MetactoHost: Michael Fauscette, CEO & Chief Analyst, Arion Research
