Agents and workflows: 14.7% on CCDV-F against 27% on CCAR-F

Agents and Workflows is 14.7% of CCDV-F — roughly 8 of 53 items — split almost evenly across three published skills: Agent Construction with Claude at 5.3%, Agent Patterns and Frameworks at 4.9%, and Agent Architecture at 4.5%. The architect foundations exam gives the same subject 27% across seven objectives. Half the weight, a third of the granularity, and a different question underneath.

Half the weight, a third of the detail

Note how flat that is. Three skills within eight tenths of a percentage point of each other means roughly two or three items each and no reason to prioritise one over another — unusual on an exam where one domain is a third of the paper and the spread elsewhere runs to thirteen to one.

Construction is the largest skill

At 5.3% this is the biggest of the three and the one least represented in the material that circulates. Its subject is how the thing is actually built: the Agent SDK against a loop you write yourself, the shape the deployment takes, and hooks as a way of making one part of the behaviour deterministic.

The examinable judgment is the same layer question the rest of the programme keeps returning to — what the supplied loop does for you, and what you gave up by accepting it. That is set out across tracks in Agent SDK, Claude Code or the API. Hooks are the sharper half: an interception point decides an outcome where an instruction only influences one.

The gate is the same on both

Agent Architecture at 4.5% covers workflow against agent, and supervision once more than one component is involved. The gate is identical to the architect version and identically disqualifying: if every step can be written down before the run starts, an agent has been ruled out and cost, tooling and familiarity never enter the discussion.

Worth carrying alongside it: the architect professional exam adds a third option below the workflow — a single well-supplied call — so the same scenario can eliminate both. That progression is covered in solution design on CCAR-P and the gate itself in when not to build an agent.

Patterns and frameworks

The middle skill at 4.9% covers tool-use loops, subagents, memory and managing the context window inside a long-running system, plus the agentic frameworks themselves. It is the skill that borders most of the rest of the exam.

Two ideas carry the weight. A subagent inherits nothing, so anything its work depends on has to be in the brief — covered in subagents and coordinator patterns. And a loop whose sole exit is a successful result has no exit at all when success is out of reach, which is why a stopping condition and a defined terminal action are part of the design rather than an addition to it.

What CCDV-F does not break out

This is the useful half of the comparison. CCAR-F names orchestration, subagent invocation and context passing, enforcement and handoff across multi-step workflows, and session state with resumption and forking, each as its own objective. CCDV-F names none of those separately — they sit inside patterns and frameworks, sharing 4.9% between them.

So a developer-track candidate revising from architect material will spend hours on session forking and handoff seams that are, at most, part of one skill worth two or three items. The reverse is worse: an architect-track candidate revising from developer material meets seven objectives having studied three skills. Neither is a scope error — the material is real on both — and both are calibration errors, which are harder to notice because nothing you read was wrong.

How to study it on this track

Two evenings, split evenly, and no more. One on construction and architecture together, because the build question and the should-we-build question are answered from the same understanding of what a loop does. One on patterns, aimed specifically at what a brief must contain and what ends a loop.

Then stop. Eight items is a real block and it is not where this paper is decided — Applications and Integration is 17 items on its own, and model selection and optimisation is nine. Agents are the most interesting material on the developer blueprint and the third-largest block on it.

Key takeaways

• 14.7% and three skills. About 8 of 53 items, split 5.3, 4.9 and 4.5 — flat enough that prioritising between them is wasted effort.
• Construction is the largest. The SDK against a hand-written loop, deployment shape, and hooks as determinism.
• The gate is unchanged. Enumerable steps rule out an agent here exactly as they do on the architect tracks.
• A brief carries everything. A subagent inherits nothing, and a loop with no stopping condition has no ending.
• CCAR-F breaks out what CCDV-F folds in. Orchestration, handoff and session state are separate objectives there and part of one skill here.
• Borrowed agent material mis-calibrates. Nothing you read is wrong; the emphasis is built for a different paper.