Daeus Jorento

Foundations: Games, Rationality, and Traps

In organizations, strategies and execution are interdependent. Outcomes depend not just on what you do, but on what others do, and what they expect you to do. Game theory begins where individual optimization ends. While formal solutions rarely exist for real-world situations, the framework is useful for pattern recognition at work.

Chess is not a game. Chess is a well-defined form of computation. You may not be able to work out the answers, but in theory there must be a solution, a right procedure in any position. Now, real games are not like that at all. Real life is not like that. Real life consists of bluffing, of little tactics of deception, of asking yourself what is the other man going to think I mean to do. And that is what games are about in my theory.
– John von Neumann

I don’t really play chess, but it’s a useful metaphor. The goal is not to find the perfect move, but to avoid traps or unwinnable positions, assuming the other player is rational. The earlier you recognize those situations, the more time and energy you save. A big part of this is recognizing which game you are actually in, rather than playing blindly.

In most environments, I assume people are acting rationally within their constraints. The useful question becomes: what rules and incentives shape their behavior?

In organizations, behavior follows rewards and incentives, not intentions. Cultures can settle into stable equilibria that no one explicitly wants. Skilled leaders design mechanisms where rules drive good outcomes naturally by making consequences, good and bad, unavoidable.

The classic example is the Prisoner’s Dilemma. Two prisoners each choose to cooperate (stay silent) or defect (betray). If both cooperate, they each serve one year. If one defects while the other cooperates, the defector goes free and the cooperator serves five years. If both defect, they each serve three years. Assuming no trust or coordination, the rational choice is to defect, leaving both worse off.

The only satisfying solution to the prisoner’s dilemma is to avoid prisoner’s dilemmas.
– William Poundstone (Prisoner’s Dilemma)

Why Organizations Struggle: Local Rationality

Failures often come not from bad people, but from locally rational decisions inside poorly designed games. If a system rewards optics, or complexity over impact, people adapt. Sometimes the same people change their behavior. Other times, different people rise to the top. Over time, different players learn how to win under different rules, and that becomes culture.

Many org conflicts are framed as values disagreements when the real issue is incentives.

Promotions

Without clear rules, promotions favor likeability and narrative skill over impact. Early in a startup, this is often fine (assuming reasonable execution). It does not scale. Over time, strong performers who missed early promotion windows get frustrated and leave.

Limited promotion slots create zero-sum dynamics, pushing teammates to compete, hoard work (e.g. take over adjacent teams’ scope), or switch teams. Leaders who encourage competition often think only about short-term achievements (‘offense’), such as moving metrics faster, and underestimate negative externalities (‘defense’), like hiding information or sabotaging peers (Superchickens Ted talk).

Rewarding impact and quality matters more than rewarding presentation, volume, or technical complexity. Incentivizing signaled competence quickly leads to overengineering. Incentivizing ideas over execution produces a lot of smart plans that go nowhere. I have personally fallen into this trap by treating company Slack like social media.

The harder problem is balancing this with principled risk-taking and teamwork, where contributions are harder to measure. I still find it useful to think of each person as a small business with a cost-benefit, measured directly through impact or indirectly through what would not happen if they left. Some criteria, like visibility, are boxes to check, not things to optimize.

If one person brings in $1M and another presents at a conference, rewarding the latter produces more conference talks.

The main trap to avoid is unclear or inconsistently enforced expectations on impact (especially for folks at higher levels). Clarity can forgive a lot of sins.

Firing

Firing the bottom 10% systematically can incentivize peer sabotage. Like running from a bear, you don’t have to be fast, just faster than someone else. It also assumes that managers can objectively identify true low performers, and not just people that they dislike (or are convenient to fire). New hires are especially vulnerable since they lack political ties. Over time, this behavior becomes part of the culture and damages the company’s reputation.

At the same time, never firing people who are safe but achieve just below minimal delivery creates a coasting dynamic. For more junior levels, up-or-out policies attempt to address this by introducing time-based quotas, but disengaging can happen whenever someone gets too comfortable and stops being challenged. Getting fired can feel personal, even when it happens for reasons beyond interpersonal conflict. Often, though, it is simply that the role is no longer needed or the fit changed. Like a subscription, it is not that I hate you, Netflix; it’s just the show I was watching ended and I no longer want the service.

Every policy has tradeoffs. An ideal structure is challenging but supportive, with reasonable pruning. The most damaging traps are unclear expectations and subjective enforcement. Making internal transfers easy helps retain strong people who are simply mismatched. Some behaviors can’t be fully prevented. People from cutthroat cultures may bring those strategies with them and need time to unlearn them.

The same incentive problems show up most clearly at organizational boundaries, where companies interact with external markets rather than just internal teams.

Hiring and Salaries: Market Games

Interviews and Hiring

Unclear overall standards, inadequate career ladders and shallow interview loops can create friend networks that inflate performance signals. A simple minimax solution is role separation: hiring managers source and advocate, while capable, rotating committees evaluate with a high-level leader approval (similar to cookie division, where one person cuts and the other chooses). Hiring managers with full autonomy past a certain scale are like referees playing on the field.

There is currently an AI arms race in job applications. Candidates mass-apply with AI, and recruiters screen with AI. This pushes companies back toward heavily prioritizing referrals. Take-home exercises tailored to the business can help screen out candidates who are mass applying and disengaged (game theory blog post). One useful interview strategy comes from the secretary problem or 37% rule: use the first set of candidates to calibrate, then hire the next candidate who exceeds that bar.

I don’t like LeetCode as a post-college SAT, but pass-fail technical screens do filter out candidates who cannot break down and solve problems in the language they will actually use, even with AI. I’ve seen data orgs without any coding or SQL screens devolve into politics as a substitute for technical skill.

When Square added a Python screen, people hired after liked it. People hired before often didn’t. They were both favoring the processes that validated them. Tenured employees value tenure. New hires value credentials. Both are understandable, and neither is unbiased.

Revamping interviews is also a way to clarify what a role actually needs and to quietly uplevel a discipline. Crowdsourcing structured interview questions helps keep both the bar and interviewers sharp.

Salaries

Matching external offers can create internal inequities. Refusing to match could tempt managers to overlevel, which erodes standards relative to the industry. Ambitious managers want to win and sometimes regret it later.

The best solutions I’ve seen combine signing bonuses with retention clauses, roles that offer real potential for impact and learning, and giving new hires agency alongside a strong support system (including enablement for relatively fast promotion when someone is slightly under-leveled but performs exceptionally). Training people well while accepting some will leave beats undertraining to retain. Room to grow is an important part of tech compensation.

On the surface, these dynamics seem isolated. Over time, these interactions start shaping how teams work, what they expect, and even what they build.

Roles and Products: Repeated Games

People unconsciously adopt roles, such as victim, persecutor, or rescuer (Karpman drama triangle). Awareness helps break these loops. Defensive reactions often recast others as attackers and justify escalation. Regular, actionable feedback reduces this. Feeling like a victim can also grant moral license to act badly, which is worth noticing in yourself.

Sometimes having a dedicated skeptic role in a group can improve decisions. People tolerate friction better when they know it’s intentional.

One surprisingly strong strategy for a repeat prisoner’s dilemma situation is a tit-for-tat. Basically ‘do no harm, take no shit’ which is a nice balance of justice and forgiveness long term.

Products often reflect org structure. Companies ship their org charts. Tight team coupling leads to cleaner product boundaries. Unclear ownership creates avoidance and turf wars, especially around risky legacy systems (e.g. ‘hot potato’ problems). Org design resembles distributed systems: balancing autonomy and coverage is hard, and incentives must match temperament.

Data teams sit at an uncomfortable intersection of these games, exposed to both agenda power and technical authority.

Data Organizations

Rick’s the right guy to evaluate the risk. He’s not the right guy to stand down the guys who want their deals done, they’d ram it down his throat.
– The Smartest Guys in the Room by Bethany McLean and Peter Elkind

Data teams need independence to stay objective and proximity to stay relevant. Under engineering, they risk becoming technically correct but not empowered or even irrelevant. Engineering leaders respect systems shipped, latency reduced, and reliability improved. They can easily under-index on counterfactual insights, strategic opportunity sizing, and recommendations on what to build. On the other hand, under business, data teams can become a decision-making function and help set the strategy early on. They can help shape strategy, prioritization, and achieve narrative ownership. However, in such a setup, they risk being more biased without guardrails. It is easier for me to tell a stakeholder their idea failed than my manager.

A good overall model achieves ‘dual alignment’, where data teams are centralized under a single strong leader with company-wide leadership influence and credible technical knowledge in at least one data domain. There could be dotted line structures into engineering in order to be closer to production pathways and infrastructure. Engineering has technical power but business has agenda power.

Metrics

For analysis, the game is speed versus accuracy. My solution is habit-building: knowledge checkpoints, early feedback, and stable reference metrics. Over time, this creates a set of facts I can move quickly with. I usually find a solid, generally accepted premise is more important than dressing or perfect logic.

Choosing the right success metric, especially in a broad ecosystem, is difficult. Metrics obey Goodhart’s Law: when a measure becomes a target, it ceases to be a good measure. For example, optimizing conversion rates in a vacuum can lead to increased friction for low-intent users. Reducing friction to enable growth increases volume but lowers conversion and can sometimes lead to higher fraud losses. The real game becomes customer identification, which is difficult, but not zero-sum.

Putting out fires, such as bug fixes, versus prevention is a stag hunt problem. In a stag hunt, everyone is better off cooperating on a high-value outcome, but each individual is tempted to chase a smaller, guaranteed win. Bugs are visible and rewarded (blog post). Prevention is silent. Without clarity, people rationally pursue short-term fixes over long-term platform health.

Leadership and Power

In peacetime, leaders must maximize and broaden the current opportunity… In wartime, by contrast, the company typically has a single bullet in the chamber and must, at all costs, hit the target.
– Ben Horowitz (The Hard thing about Hard Things)

In peacetime, leaders still need to stay grounded in business impact. Otherwise, they drift toward employee satisfaction as the primary signal of success. Those metrics often move with company performance rather than leadership quality, especially in public companies where compensation is tied to stock. Managers who see their role as keeping their team happy will struggle to challenge them or adapt as conditions change. A successful leader must ask at the start of each day: How can we do more to improve delivering positive impact for our customers?

Once, when I was struggling with my manager, I asked a department head for help moving roles. They said they couldn’t create one for me. That clarified something for me: leaders have to operate on principles, not favors. It helps to ask what a company would look like if everyone worked this way. Favor-based help creates dependence and slowly erodes meritocracy. I have seen managers defend someone during layoffs only for both to be cut later because they became a package deal.

Leaders generally need to support and enable their managers unless there is clear evidence of poor performance. Patterns outweigh anecdotes. Reduced team impact and productivity is often a more reliable signal of problems than individual complaints (or even turnovers).

Ambitious middle managers often expand scope through headcount to advance. This is sometimes labeled ’empire building’, but it is usually a rational response to the incentives in place. Left unchecked, it can balloon orgs without increasing impact. Measuring impact per headcount or treating orgs like businesses with an ROI helps constrain this. Setting clear OKRs at each team / org level and enabling regular business reviews that track progress can also be effective. Platform teams are harder to evaluate, but can be approximated by comparing alternative vendor costs, volume processed, or outcomes in a world where the team does not exist. Managing orgs at scale starts to resemble portfolio management.

Opening new opportunities reduces zero-sum pressure between strong leaders. This is why the skill of creating new surface area is so valuable when properly harnessed. When things feel stuck but you want to stay, it can be more productive to open a new business line that can fund itself, much like a startup. The constraint is staying grounded in real customer value, not just winning by getting funded (Paul graham article).

Solutions: Designing Better Games

Recognizing zero-sum games and finding ways out of them is one of the most valuable skills in an organization. In product ecosystem shared spaces, this often means bundling and complementing products rather than letting teams compete internally.

People need skin in the game and proximity to the consequences of their decisions. Clear rules applied consistently matter more than perfect judgment. Clarity, humility, and effort can forgive many mistakes.

If I could avoid one trap, it would be unclear hiring standards, which let unproductive but likeable hires slip through. Closely related is the lack of objective leveling and promotion criteria that tie rewards directly to impact rather than visibility or competency signaling. At the team level, regularly showcasing good work helps prevent strong contributors from quietly building resentment when their efforts go unseen.

Optimize for the game actually being played. Growth requires different behavior than reliability. Local maxima strategies, like short-term cutthroat moves, may feel like success to some but prevent reaching a global maximum. Reaching a global maximum requires more fair play and everyone rowing the same way toward real impact. Not everyone is rational, but incentives compound. Companies can outgrow their problems for a while, but they surface eventually. A simple check is to ask how things would turn out if everyone acted the same way.

Thanks Rob Wang for helping with review!