A Strong Technical Team Is More Than Its Hero Leader
Daniel Dennett’s account of distributed competence gives technical leaders a useful way to examine their teams.
A large organisation may act competently even when no single person understands the whole system. Its capability is spread across people, interfaces, routines, tools, feedback and accumulated decisions. The public story may still credit one exceptional leader, but the operating reality is usually more distributed.
This matters for hiring, onboarding and succession. If a team works only because one person continuously interprets every situation, the team does not possess that competence. It is borrowing it.
The hero can hide a weak system
Hero leadership often looks efficient in the short term. One experienced engineer knows where the risks are, resolves ambiguous incidents, translates between departments and remembers why old constraints exist.
The problem appears when the person is unavailable or overloaded. Decisions slow down, reviews become ceremonial and everyone waits for the same source of certainty. The organisation may have many capable people, but it has not turned their capability into a dependable operating system.
The practical question is not whether the leader is brilliant. It is whether the team can preserve the right obligations when the leader is absent.
Distribute competence through visible mechanisms
Distributed competence does not mean removing ownership. It means giving local actors enough information and authority to act while preserving the boundaries that keep the whole system safe.
A team needs at least five visible mechanisms:
- A shared obligation. People know what outcome the system must protect, not only which ticket they received.
- Local decision boundaries. An engineer can tell which choices belong to the role and which require escalation.
- Executable feedback. Tests, monitoring and user signals reveal whether an action produced the expected result.
- Decision memory. Important assumptions and exceptions remain accessible after the meeting or author disappears.
- Recovery ownership. Someone is responsible for observing the result and correcting it when reality differs from the plan.
These mechanisms allow the team to coordinate without pretending that every participant has a complete model of the organisation.
Hiring for a distributed system
A hiring process often rewards individual explanation: the candidate who presents the cleanest architecture, speaks most confidently or recalls the largest number of patterns.
Those signals are useful but incomplete. Distributed work requires another capability: making local expertise usable by other people.
The candidate should be able to:
- state the assumption behind a recommendation;
- identify the evidence that would change the recommendation;
- leave a decision that another engineer can inspect;
- distinguish a reversible experiment from a consequential commitment;
- ask for help before crossing a boundary they cannot adequately verify.
This is not a preference for cautious or quiet engineers. It is a preference for work that remains reliable outside the individual performer.
Onboarding should test the team as well
When a new specialist struggles, organisations often treat the problem as an individual deficit. Sometimes it is. But onboarding is also a test of the system receiving the person.
Can the newcomer find the current source of truth? Are decision boundaries explicit? Do tests correspond to real obligations? Can the person observe the result of a change? Is there a named owner for an exception?
If the answer is no, adding more introductory meetings will not solve the underlying problem. The team is still depending on oral reconstruction by insiders.
A useful onboarding task therefore has two outputs:
- a verified change made by the newcomer;
- one improvement to the mechanism that made the change possible.
The second output may be a test, an ownership rule, a corrected runbook or a clearer interface. This turns onboarding into a gradual reduction of hidden dependency.
Leadership becomes architecture
In a distributed technical system, leadership is not the continuous production of answers. It is the design of conditions under which good local decisions can accumulate.
The leader still sets direction, resolves genuine conflicts and accepts responsibility for consequential choices. But the leader should not be the only route through which the organisation can think.
A mature team can continue operating when the leader is offline because its obligations, feedback and authority do not disappear with one person’s attention.
That is a more demanding standard than celebrating a hero. It asks whether the organisation has converted individual competence into a system that other people can safely join, inspect and improve.
Source episode: Why Can a System Be Smarter Than Its Leader? | Daniel Dennett
Top comments (0)