Video summary

Unfortunately, I Was Right

Main summary

Key takeaways

Technology

Summary of Subtitles (Tech/Product Concepts + Predictions)

  • Token-cost budgeting will become a competitive constraint (token efficiency).

    • The speaker mocks “token maxing” (using unlimited tokens) and argues companies will reverse course because costs won’t scale.
    • They reference a prediction that came true quickly (about 8 days) related to AI agents and token spending becoming a major issue.
    • They attribute the shift to basic corporate incentives: budgets get scrutinized like hardware/resource requests.

  • Costs likely drive new spending rules and “token governance.”

    • Companies will push for “token efficient” approaches instead of allowing teams to spend tokens freely.

Predictions

Prediction 1: Trading tokens for incentives and/or equity; open-source “token donations”

  • Some AI labs may trade tokens for equity (the speaker says it’s already happening).
  • A “token donation” model for open source is predicted:
    • Users donate large token budgets to cover CI/automation costs for open-source projects.
  • Framed as analogous to donating compute (e.g., protein-folding-style), but with tokens instead.

Prediction 2: Token stipends per employee (token budget tied to bonuses)

  • Employees receive a yearly token stipend.
  • If they use fewer tokens than their budget, they earn a bigger bonus.
  • Example dynamic: “blow the budget vs. be efficient” influences compensation.

Prediction 3: “Token Agile” / “token poker” replacing planning methods

  • The speaker imagines Agile-style estimation evolving into a game-like approach:
    • Teams estimate task cost in tokens using “cards”/guesses.
  • Expected outcome:
    • A consulting-friendly “new process,” effectively turning planning into token-cost forecasting.

Prediction 4: Organizational/team token budgets create managerial overhead + prompt optimization culture

  • Instead of one company-wide AI budget, there will be org-wise and team-wise token budgets.
  • Likely outcomes:
    • Middle managers focused on negotiating/petitioning for token allocations
    • Teams “pair prompting” (collaborative prompt crafting) to reduce token usage
    • More “reviewed” kickoff prompts for long-running agent tasks (multi-hour/day runs)
  • Prompts become artifacts discussed like GitHub contributions—teams debate context/prompt strategy to save tokens.

Prediction 5: Rewarding “highest AI usage” could distort engineering incentives

  • The speaker predicts companies may reward the most AI usage by giving more budget to those perceived as top performers.
  • Example mechanism:
    • Auditing logs where more code/output equals more tokens; less output equals fewer tokens.
  • Connection to “100X organization” concepts (from ClickUp):
    • Higher output requires larger budgets, concentrating resources.
  • Extreme risk:
    • Organizations could be dominated by “slop cannons” (very high production/low quality AI output) and fail under the burden.

Main Speakers / Sources Mentioned

  • Sam Altman (referenced regarding token-cost concerns)
  • George Hotz (quoted regarding AI agents adoption as a costly mistake)
  • Uber COO (referenced in the context of “token maxing”)
  • Microsoft (referenced via “Microsoft’s org chart” as an example of structure)
  • ClickUp (referenced via “100X organization” concept)
  • Cursor / Composer 255 (sponsor mentioned; Cursor as the source)

Sponsored Mention

  • Composer 255 (from Cursor)
    • The speaker promotes it for:
      • fast back-and-forth interactivity
      • strong output quality
      • affordability for research/debugging without “breaking the bank”

Original video