CLAUDE CODE AGENTS: ADVANCED ORCHESTRATION - Chapter 9, Exercise 1
Faster in Wall-Clock Time, Not Necessarily Cheaper Overall
====================================================================================

QUESTION:
Explain why running five agents in parallel is faster in wall-clock time
but not necessarily cheaper overall than running the same five tasks one
at a time.

SOLUTION / EXPLANATION:

Wall-clock time and total cost are two genuinely different measurements,
and parallel execution improves one while leaving the other largely
unchanged - or even making it comparatively less efficient, depending on
overhead.

Running five independent tasks sequentially means the total wall-clock
time is roughly the sum of all five tasks' own individual durations, one
after another. Running the same five tasks in parallel instead means the
total wall-clock time is roughly just the duration of the SLOWEST single
task among them, since all five run at the same time rather than one after
another - this is exactly the wall-clock benefit Chapter 3 described.

But total cost is a different measurement entirely: it reflects the sum of
resources actually consumed across all five tasks, regardless of whether
they ran one after another or all at once. Each of the five tasks still
requires its own full amount of work - its own model calls, its own tokens
processed - whether it happens sequentially or in parallel with the
others. Running them in parallel doesn't reduce how much total work each
individual task requires; it only changes how that work is scheduled in
time. Five tasks each costing a certain amount still cost roughly that same
total amount added together, whether they're spread out over time
sequentially or compressed into the same time window in parallel.

So parallel execution is a genuine tradeoff, not a strictly better choice
in every sense: it buys real, favorable wall-clock time at the cost of
concentrating (not reducing) the same total resource expenditure into a
shorter window, rather than actually lowering that total cost.

--------------------------------------------------------------------------
WHY THIS WORKS AS AN ANSWER:
It distinguishes wall-clock time from total cost as separate measurements,
explains specifically why parallel execution improves the former (shortest
duration is the slowest task) without reducing the latter (each task still
consumes its own full resources), and frames this as a genuine tradeoff
rather than parallel execution being simply better.