If I open Instructions for Claude and find the content doesn't quite match what I remember configuring in Global Instructions, what should I do?
Don't assume a system error first. The more likely explanation is that your Cowork-specific Global Instructions and whatever instruction settings you already had in other contexts (regular Chat, say) got folded into the same Block during the merge, and the resulting combined or interleaved content naturally looks different from what you remember as "just the Cowork settings" on their own — that doesn't necessarily mean data was lost, it means two previously separate setting sources now share the same display location.
The more practical approach is reading through the full content currently shown from start to finish, confirming line by line whether each instruction is still something you want to keep, rather than rushing to compare it against "does this match the old version I remember." If you find some instructions no longer apply or conflict with your current working habits, editing this now-unified setting block directly into what you actually want today is more efficient than trying to reconstruct the exact pre-merge old version.
Does the list of folders shown under Trusted Folders mean Claude can still access everything in those folders right now?
That list represents the granted access scope — the local folders you previously agreed to let Cowork access — and it carries over into the new experience remaining in effect. There's no indication the merger itself automatically revoked or narrowed your original grants. But it's worth noting that the existence of an access grant doesn't mean you still clearly remember why you opened up a particular folder in the first place, or whether what's currently in that folder still matches the context you originally granted access under.
The more solid approach is treating this migration as an opportunity to revisit your access scope — going through each entry listed under Trusted Folders and asking yourself whether that folder still needs to stay open, and whether anything sensitive has since been added there that you wouldn't actually want Claude touching. Access scope doesn't automatically become more cautious or more permissive just because the interface changed — it always reflects a decision you made at some past point, and it's worth periodically checking whether that decision still makes sense today.
"Memory from local tasks stays inside the original task" — does that mean I can't apply memory accumulated in one local task to a new task afterward?
According to Anthropic's documentation, memory accumulated within a local task does stay bound to the task that originally ran it — it isn't pulled out into a general memory that applies across all new tasks. That means if you want a working habit or preference developed within a local task to also be remembered by a new task you open later, there's currently no automatic mechanism that extends it for you — you'd effectively need to re-establish that context in every new task.
If certain working habits or preferences are things you want to carry across tasks long-term, the more practical move is not relying on local task memory to automatically extend, but instead writing those preferences explicitly into the "Instructions for Claude" setting — because that setting applies to every conversation rather than being bound to a single task, it's the actual channel under the current architecture that genuinely carries across tasks. Memory within a local task is better suited to continuity within that specific task's own context, not as a storage location for long-term preferences.
With the Web Search toggle gone and Claude now deciding on its own whether to search, is there any way to actively turn off search for a specific conversation if I don't want it used?
Anthropic's documentation on this point focuses on "Claude now searches automatically when it judges that searching would help," without separately mentioning a switch that lets you manually disable search. That means control has shifted from "you decide whether to search" to "Claude judges based on the task's content" — if you simply want a particular task to avoid web search entirely, there's currently no direct equivalent to the old toggle's function.
If avoiding web search is an important consideration for a particular task (you want Claude relying entirely on the material you provided, say, or the task involves sensitive content you'd rather not have cross-referenced online), the more practical approach is stating this explicitly in your instructions — telling Claude directly that this task doesn't need to look anything up online and should work only from what you've provided. There's no one-click button in the interface for this, but expressing your need clearly through the instruction itself still lets Claude adjust its approach accordingly.
If your account has switched to the merged Cowork-and-Chat experience, the first time you open the interface you'll likely notice a few familiar setting entry points are gone — that doesn't mean your configuration data was lost. Anthropic's documentation explicitly lists several things that have specifically "moved house," and this piece maps out where each key Cowork setting now lives in the new interface, so you don't have to hunt for them by trial and error.
The Cowork tasks you previously created now appear alongside your regular chats in the Recents list, no longer in a separate, dedicated tab. That means if you want to find an old Cowork task, browsing Recents directly will show it — you don't need to switch to some dedicated "Cowork" area first to see them. Anthropic's documentation also notes these older tasks open exactly as they did before, so you can pick up right where you left off — the interface moving didn't change the task's content along with it.
This is the change most likely to confuse existing users. What was previously the "Global Instructions" setting in Cowork is now folded into "Instructions for Claude" under Settings > General. Anthropic's documentation includes a specific note worth taking seriously: check that your instructions are what you want. The implication behind that note is that this isn't simply a relocation — your global instructions, previously configured specifically for Cowork, are now merged into the same setting Block as instructions from other contexts. Whether the content survived the merge fully intact, and whether it conflicts with or gets overridden by other existing settings, is genuinely worth opening up and checking yourself, rather than assuming it carried over seamlessly.
If you're in Claude Desktop, the folder where Cowork used to save your work still exists — you'll find an item called "Storage folder" in Settings. Meanwhile, the local folders you previously granted Cowork access to are now listed under "Trusted folders." These are two separate concepts — Storage folder is where Claude's output gets saved, while Trusted folders is the list of source folders you've authorized Claude to read from. Anthropic's documentation presents them as separate setting items rather than merging them into one generic "file settings" catch-all. If you can find only one and not the other, it's worth remembering these are two different entry points, not two names for the same setting.
If a Cowork task you previously ran was entirely local — running only on your own computer — the memory accumulated from that task stays with the task itself and doesn't get pulled out and moved elsewhere. That means this particular slice of memory isn't something you can manage centrally within the new experience — its home is still tied to whichever conversation originally ran the task. If you're used to telling Claude about your working habits or preferences and expecting it to remember, and that interaction happened inside a local task, that memory's current scope is still bound to that specific task, rather than becoming general memory applicable across every context.
If your interface previously had a dedicated "web search" toggle, it's gone after this merger too — but Anthropic's documentation is clear this doesn't mean web search disappeared; Claude now searches automatically whenever it judges that searching would help, with no need for you to flip that switch manually. This differs slightly from the earlier items where "things moved elsewhere" — Web Search specifically had its manual control interface removed entirely, replaced with Claude judging on its own whether to search.
If you used to launch deep research through a specific entry point, there are now two ways to trigger it: typing /deep-research directly, or clicking the "+" button below the chat input and selecting "Research." That means what might previously have been a persistently visible button now requires you to actively type a command or click through a menu instead. If your workflow relied on this feature being highly visible and easy to reach at a moment's notice, this change may take some time to build a new habit around.
If your account just switched to the new experience, the more practical first move isn't rushing into a new task — it's spending a few minutes checking each of these entry points against what's covered here: open Settings > General to confirm Instructions for Claude is intact and complete; find Storage folder and Trusted folders in Settings to confirm what's shown matches the access scope you remember granting; check Recents to confirm your old Cowork tasks are all still there and pick up normally. This check doesn't take long, but it buys you the chance to catch any gaps that might have surfaced during the migration early, rather than discovering a setting doesn't match what you expected only when you actually go to use it.