Projects and connectors: the product half of CCAO-F, and only CCAO-F

Four CCAO-F objectives name Claude product features: output formats including artifacts, selecting among product surfaces, configuring Projects with instructions and knowledge sources, and managing uploaded knowledge and connectors. Check the other three blueprints and none of them names a single one. The two domains carrying most of this material are 12% each — roughly 14 of 60 items between them.

The only track that tests the product

This is the clearest way to see what separates the associate track from the rest. The developer and architect exams treat Claude as something you build against; this one treats it as something you use, configure and put in front of colleagues. Anyone moving up a track should expect this material to disappear entirely — see the certification chooser for how the scopes differ.

Choosing a surface for a job

The objective names selecting appropriate product features, and the discriminator is the shape of the work rather than the richness of the surface. A question with a one-sentence answer needs no container, and putting it in one is overhead nobody asked for. Something that will be revised and handed to someone else is a work product. Work that recurs, with material several conversations need, wants a place where that material persists.

The recognisable wrong answer reaches for the most capable surface regardless of the job. It is not wrong in the sense of failing; it is wrong in the sense the exam cares about, which is that the decision was not made against anything.

What a Project actually is

Two things, per the objective: instructions and knowledge sources, held in one place so that several conversations start from the same footing. That framing is more useful than any feature description, because it says what the thing is for.

Which leads to the examinable judgment: what belongs in the persistent layer and what belongs in the conversation. Anything every conversation needs — the standard, the audience, the material — belongs where it persists. Anything specific to today belongs in today’s conversation. Getting that backwards produces either a container nobody benefits from or one so crowded that its instructions compete with each other, which is the same crowding examined in Claude Code configuration on a different track.

Connectors: reach is not permission

The connectors objective sits next to a governance domain worth 15%, and the two meet on one point. Connecting a source makes material reachable. It does not establish that the material may be used for the purpose you have in mind.

Scenarios in this family are built on exactly that gap: the integration works, the data is right there, and the correct answer is about whether policy permits this use rather than whether the connection succeeded. It is the same judgment covered in governance and responsible use, arriving from the product side.

Restart, summarise, or persist

One objective in the product-selection domain covers context limits and memory, and names the three moves directly: when to restart, when to summarise, and when to persist. They are different answers to different situations rather than degrees of the same one.

Restarting is right when the accumulated history has stopped helping. Summarising is right when the conclusions matter and the working does not. Persisting is right when something has to survive beyond this conversation entirely — and the tell is that it will be needed by a conversation that has not happened yet. The reliability version of the same problem is in context management.

The one domain that dates

A caution worth stating plainly. Product surfaces change far faster than exam blueprints, and the official guides are versioned documents — the current set is Version 1.0, effective July 2026. Any study material describing where a setting lives or what a menu offers is describing a moment.

So prepare this domain by judgment rather than by feature list. Which surface suits which shape of work, what a connector establishes and what it does not, what belongs in a persistent layer: those survive a product update. Screenshots do not, which is also why the material here stays deliberately general and points at the Anthropic Partner Academy certification pages for anything current.

Key takeaways

• Only CCAO-F names the product. Projects, artifacts and connectors appear in four of its objectives and in no other blueprint.
• Roughly 24% across two domains. Product and model selection, plus configuration and knowledge management, at 12% each.
• The surface follows the work. How the output is consumed, and whether anything must survive the conversation.
• Reach is not permission. A working connector establishes what is possible; policy decides what is allowed.
• Restart, summarise and persist differ. Three answers to three situations, not three strengths of one.
• Learn judgment, not menus. Product detail dates faster than the blueprint that references it.