The Twenty Minute VCDaniel Dines on Why Work Processes Not Models Will Be The Most Valuable Asset in AI
At a glance
WHAT IT’S REALLY ABOUT
Enterprise AI value will accrue to workflows and governance, not models
- Daniel Dines argues that the lasting value in enterprise AI will accrue to the ‘Map of Work’—the detailed, company-specific workflows, exceptions, and governance—rather than to any single foundation model.
- He distinguishes AI ‘memory’ from true ‘learning,’ claiming today’s models don’t transform through experience the way humans do, which forces enterprises to document their operating reality for AI to be effective.
- He explains why probabilistic AI is unreliable for long, multi-step execution and proposes a pattern where AI builds and fixes deterministic automations that are testable, auditable, and exact.
- He warns that AI-driven reorganizations can backfire unless companies account for the hidden relational and cultural outputs people provide beyond their measurable job descriptions.
- He predicts enterprise inference will primarily run on cheaper models with strong model-switching capability, driven by cost, governance, and fear of vendor lock-in and IP leakage.
IDEAS WORTH REMEMBERING
5 ideasWorkflows—not models—become the enduring moat in enterprise AI.
Dines argues that frontier LLMs are largely “interchangeable” over time, while an enterprise’s codified processes, exceptions, approvals, controls, and tool integrations are unique and defensible. The durable asset is the company-specific workflow layer (“Map of Work”) that lets you apply whichever model is best this quarter without losing operational know-how.
AI ‘memory’ isn’t the same as human learning, so enterprises need manuals.
He distinguishes “memory” (notes, retrieved context, scratchpads) from “learning” (weight updates / being changed by experience). Because models don’t truly learn on the job in the way humans do, enterprises must externalize knowledge into documented procedures and exceptions for AI to execute reliably.
Probabilistic agents struggle with exactness at scale; deterministic rails matter.
Dines highlights that probabilistic error compounds across long, multi-step processes, making pure agentic execution risky for enterprise operations. The right pattern is for AI to call deterministic tools (software, automation, calculators) for exact steps, using AI primarily to design, diagnose, and repair workflows.
AI will ‘print’ enterprise software faster than it will safely run enterprises autonomously.
He sees an “asymmetry”: it’s getting easier to use AI (coding agents) to build automation than it is to safely deploy autonomous agents to run core business processes end-to-end. Enterprises can audit, test, govern, and lock down software/automations in ways they cannot with free-roaming agents.
Workforce transformation requires mapping hidden human value—not just cutting headcount.
Dines cautions against blanket layoffs justified by AI, because many roles carry hidden value: customer trust, cultural stewardship, mentoring, and initiative. He proposes a “ledger” of visible and invisible outputs to decide what to automate, what to augment, and who to retain or redeploy.
WORDS WORTH SAVING
5 quotesModels are interchangeable, but the workflow, the map of work and the workflows around the map of work is where the real value is.
— Daniel Dines
AI doesn't alter its weights on the job in the way humans are transformed by a job.
— Daniel Dines
It has become very clear to me that, um, another limitation of AI is what I call exactness.
— Daniel Dines
You need to hand the Map of Work to AI in order to be successful.
— Daniel Dines
If the work at the quality were better of a human can be done by a machine, I will hire today a machine even if it's more expensive man- than a human.
— Daniel Dines
High quality AI-generated summary created from speaker-labeled transcript.
