Economics and Control Are the Same Conversation Twice
Feld and Mendelson show that liquidation preferences, participation rights, and anti-dilution provisions are all arguments about who captures value when things go well or badly. An AI-native company has an analogous internal structure: agents capture the operational output, but humans capture — or should capture — the accountability for how that output was produced. Founders who give away board control before they understand this parallel are negotiating both deals at once, usually without realizing it.
The Irreversible Decision Belongs to a Human
Venture Deals draws a hard line around protective provisions — the clauses that give investors a veto on major decisions. The logic is that certain moves cannot be undone. The same logic governs agent design. Reversible, well-specified, measurable work can run autonomously with fast feedback loops. Wire transfers, signed contracts, and public commitments cannot. Approval layers should match the stakes: not every action needs a human, but every irreversible one does, and that threshold belongs in the architecture before you raise, not after.
Trust Is Designed, Not Assumed
The book's deeper lesson is that a term sheet is a trust document written in advance for a relationship that will be tested under pressure. Investors who seem aligned in good times reveal their actual incentives during a down round. Agents reveal their actual failure modes under novel inputs. In both cases, the founder's job is to design the escalation path — board consent mechanics in one domain, eval harnesses and human-in-the-loop checkpoints in the other — before the pressure arrives. Knowing the game before sitting at the table applies equally to a cap table negotiation and a production incident.
