Principle of Least Commitment

Deciding every detail early can make learning expensive

Two small teams are each building a stair-climbing inspection robot.

The first team wants a complete plan before fabrication begins. It chooses the wheel diameter, climbing mechanism, battery position, motor, enclosure and control layout. Each choice makes sense on its own, and together they make a coherent design.

The first prototype climbs the workshop steps. On a steeper staircase it tips backwards.

Moving the battery would improve its centre of mass, but the enclosure and motor mount were designed around the original position. A different wheel mechanism would help, but its interface doesn't fit the frame. The fabricator waits with parts that may no longer be useful while an engineer redraws assemblies that had already been approved.

The plan locked in more detail than the available evidence could support.

The second team starts by fixing a smaller set of things:

• The robot must remain stable on the specified range of stairs.
• It must fit within a maximum width and weight.
• Every mechanism must connect through the same structural and electrical interface.
• Loss of power must leave the robot in a safe state.
• A shared rig will test traction, tipping, current draw and recovery.

The second team develops each wheel mechanism and battery position only far enough to run the next test. Each open choice has a rule for narrowing it and a date for deciding.

Evidence from the rig eliminates two mechanisms. The team then commits to final tooling with fewer assumptions to undo.

This is the Principle of Least Commitment:

Make the decisions needed to keep the work moving, and leave choices open when later evidence can settle them more cheaply.

The goal, safety limits, interfaces, next action and decision conditions are commitments too.

What needs deciding now?

A decision deserves early commitment when at least one of these is true:

• Other work cannot proceed without it.
• Delay would make the decision materially more expensive.
• A shared interface or safety boundary requires coordination.
• The choice is cheap to reverse but costly to leave ambiguous.
• No useful new information will arrive before the deadline.

A decision deserves deliberate delay when:

• Several options still satisfy the current constraints.
• A near-term experiment will distinguish them.
• Reversal becomes expensive after implementation.
• The choice affects many later decisions or pieces of work.
• Keeping the options alive costs less than undoing the wrong one.

A delayed decision should have:

1. A reason it remains open.
2. The information expected to resolve it.
3. A cheap way to preserve the live alternatives.
4. A trigger or date at which a choice must be made.
5. An owner with authority to make it.

Without those five things, the team is putting off a decision rather than preserving a useful option.

A partial plan can still be clear

Suppose a project contains four actions:

• Build a test fixture.
• Order standard fasteners.
• Test the first mechanism.
• Complete the final assembly.

The fixture must exist before the test. The test and fasteners must both be complete before final assembly. The fixture and fastener order can happen in either order.

A fully ordered plan still forces one sequence:

Order fasteners → build fixture → test mechanism → complete final assembly

A least-commitment plan records only the dependencies that matter:

• Build the fixture before the test.
• Test the mechanism before final assembly.
• Order the fasteners before final assembly.

The team can respond to supplier lead times without pretending the plan changed, because the order between buying fasteners and building the fixture was never a real dependency.

The same distinction appears in architecture. A product may need one durable definition of “settled payment” without choosing a payment provider forever. A data system may need a record that cannot be altered without deciding today which future reports will use it.

Name the dependencies that let people work independently, and leave unrelated ordering decisions open.

Where delayed binding is useful

Planning systems

Least-commitment planners record only the order that matters and choose details only when a conflict forces a choice. If two steps can happen in either order, the planner keeps both routes open instead of inventing a full sequence too early.

Daniel Weld's 1994 introduction to least-commitment planning explains how partial-order planners developed. The practical idea is simple: keep several workable plans, then add a constraint only when a dependency or conflict requires it.

Software architecture and engineering design

Late binding means choosing a specific implementation only when the system knows enough to choose it. Stable interfaces let teams change a storage system, outside provider or algorithm without rewriting the part that defines what the product does. Set-based engineering does something similar for physical designs: it keeps a small group of workable designs alive till tests show which trade-offs are real.

Stable interfaces still need evidence. The robot team tests three likely mechanisms and confirms the shared interface will fit the likely winner before committing to it.

Law and organisational design

A constitution can commit to durable rights, decision authority and amendment rules without spelling out every future procedure. Later law and policy can respond to situations its authors could not predict. In an organisation, a leader can define the outcome, spending limit and escalation conditions while leaving the informed team free to choose the method.

Leaving a rule open can give the most powerful person informal discretion while everyone else waits. Name who can decide, when the decision is due and how someone can appeal it.

In each case, commit the constraints people need for coordination and delay only the choices that later evidence can resolve.

What to decide now

Use the model when:

• A decision is expensive to reverse.
• Important evidence will arrive soon.
• Several options currently satisfy the requirements.
• Teams are planning later work around details that are still guesses.
• A roadmap presents one exact sequence despite weak dependencies.
• An architecture is becoming generic for hypothetical futures.
• You are keeping everything open but can't name the next action.

Write down the outcome, constraints, shared interfaces, next test and the choices that remain open:

1. Purpose: What outcome are we committing to produce?
2. Invariants: What must remain true in every acceptable solution?
3. Interface: Which boundaries must be stable so work can proceed independently?
4. Next action: What concrete step creates value or information now?
5. Later choices: Which decisions remain open, what evidence will resolve them and when?

This list keeps the next step concrete without fixing implementation details the team hasn't tested yet.

Keeping options open has a cost

Parallel prototypes and flexible interfaces consume money, attention and test effort.

The same choices add components and test cases, and a team carrying five plausible architectures may produce less evidence than one team completing a strong first version.

Early commitment can also make coordination easier, lower costs as the same approach is repeated and let expertise build over time. If no decision-changing evidence will arrive, waiting simply delays those benefits. Research comparing planning strategies has found that partial-order, least-commitment approaches aren't always more efficient than earlier commitment. A NASA analysis of commitment strategies makes that limitation explicit.

A team can keep several architectures alive without testing any of them, so each delayed choice needs a dated experiment or decision trigger.

Require every delayed choice to earn its delay:

• Which future evidence changes the decision?
• When will that evidence arrive?
• What does preservation cost?
• What happens if the deadline arrives without a clear winner?

If those questions have no answer, choose a workable default and learn from operating it.

How this differs from Real Options and Least Power

Real Options treats flexibility as an asset: a small investment buys the right, without the obligation, to make a larger commitment after uncertainty resolves. The Principle of Least Commitment is a planning discipline. It structures the work so unnecessary choices remain open while required work continues.

The Principle of Least Power says to use the least expressive representation capable of solving the problem, because simpler languages are easier to analyse and constrain. The Principle of Least Commitment concerns timing: decide only what the current constraints require, and bind the remaining choices later.

Least Power limits what the chosen tool can express. Least Commitment delays choices that current constraints don't yet require.