Matt Fitzpatrick: Who Wins the Data Labelling Race & Why Al Needs Forward-Deployed Engineers

Enterprise AI Adoption Reality

• MIT reports only 5% of Gen-AI deployments are currently working in any form.
• Gartner predicts 40% of enterprise Gen-AI projects will likely be canceled by 2027.
• Externally driven AI builds are 2x as effective as internal enterprise builds.
• Enterprise adoption faces a "cognitive dissonance" where model performance has risen 40–60% in two years, yet trust and deployment infrastructure lag.
• Successful enterprise AI requires model risk management, validation, and observability processes similar to those used in banking credit models.

Failures of Internal Build Strategies

• A major e-commerce retailer spent $25 million building a custom return-handling agent but shut it down months later.
• The retailer reverted to a deterministic flow because their custom evaluation metrics failed to detect high-risk hallucinations (e.g., granting $2 million refunds).
• Internal builds often lack the discipline regarding ROI, milestones, and timelines that vendor contracts enforce.
• Enterprises frequently struggle to hire top-tier AI engineers, with talent largely concentrated in startups and large tech firms.
• A shift toward open architecture (e.g., MCP, voice agents) is challenging the desire for fully internal, proprietary AI stacks.

Forward-Deployed Engineering (FDE) Model

• Invisible does not charge for forward-deployed engineers; clients pay only when software passes user acceptance testing and works.
• The FDE motion is distinct from traditional solutions engineering; it involves configuring core platforms to build hyper-specific workflows in 2–3 months.
• Ongoing fine-tuning is required for AI solutions, particularly as market contexts change (e.g., new GLP-1 drug launches in healthcare).
• The business model rejects the "SaaS out-of-the-box" paradigm for complex AI, aligning instead with how Machine Learning has always been sold in enterprises.
• Invisible operates 8 global offices with 450 people, focusing on in-person collaboration to build culture and trust.

AI Training and Data Dynamics

• The belief that synthetic data will replace human feedback is considered a major industry misnomer; human-in-the-loop is required for complex reasoning tasks for the next decade.
• Invisible's "Expert Marketplace" has 1.3 million active experts, sourcing specialized talent (e.g., 17th-century French architecture) within 24 hours.
• Pricing for data expertise is dynamic and context-dependent, similar to Uber's surge pricing, rather than fixed.
• The market for AI training is not moving toward a single dominant player; instead, it is evolving toward a fragmented landscape of 3–5 major specialized players.
• Revenue concentration remains high in the sector, but Invisible sees growing diversification across sectors like legal, healthcare, and agriculture.

Strategic Shifts and Investment Decisions

• Invisible raised $130 million in capital to invest heavily in technology rather than prioritizing immediate profitability.
• CEO Matt decided against "harvesting" capital now, betting on a 10–20 year growth horizon in the current high-growth environment.
• Strategy in the AI world is described as "overrated" compared to rapid iteration; the business model requires adapting to tech shifts every 3 months.
• The company is shifting from a remote-first model to co-location in key hubs (NYC, SF, London, Paris, D.C., Austin) to enhance culture and execution.
• Matt believes "fake it till you make it" is dangerous in non-deterministic AI systems; transparency regarding capabilities is prioritized to build trust.

Talent, Culture, and Hiring Philosophy

• The CEO advises startups to hire "all-around athletes" rather than rigidly role-specific employees, allowing staff to rotate across 5–6 different functions.
• Culture must be enjoyable and intellectually challenging; "brutal cultures" are incompatible with the research and exploration required in AI.
• Leadership should move toward flat hierarchies, empowering teams at the edge to make decisions based on consistent values and tooling.
• Recruiting is identified as the primary CEO responsibility: finding great people, creating a culture they love, and building something that makes them wealthy.
• Matt references a "Saban principle" for hiring, emphasizing that great institutions are built on recruiting the best players, not just processes.

Forward-Looking Outlook and Industry Optimism

• Matt predicts the most significant ROI from AI will come from AI-native businesses that disrupt physical world services (e.g., loan servicing, tax accountancies) rather than traditional SaaS.
• He is optimistic about AI's environmental impact, noting data centers currently represent <1% of global electricity, with AI-driven grid optimization offering net positive outcomes.
• Healthcare presents a major opportunity for cost reduction, with AI capable of cutting the $14,000 per capita U.S. spend by improving administrative efficiency and reducing diagnostic errors.
• Education is identified as the most transformative sector, where AI can democratize access to learning and bypass the failing traditional K-12 and university debt systems.
• The CEO anticipates that 99% of enterprise AI adoption will depend on specific task precision and trust, not general model benchmarks.