19 November 2026  ·  The honest list

What AI Can't Do in Standard Time® (and Why That Protects You)

When you hear that an AI assistant can work with your shop data, the first question should be: what can it break? Here is the plain answer. Claude and ChatGPT can look things up and, after you approve a preview, change project tasks. There is a lot they cannot touch, and we built it that way on purpose.

Six doors: five locked for delete or merge, time logs, expenses, invoices, and existing project settings, and one open for task changes after you approve

Five doors stay locked. One opens, with your approval.

The Short Version

It cannot…So you stay in charge of…
Delete or merge anythingRemoving records
Edit time logs, expenses, or invoicesPayroll and billing records
Change existing project settingsHow projects are set up
See more than your login can seeWho sees what
Save anything without your approvalEvery change

The Limits, One by One

Red marks something the assistant cannot do at all. Orange marks a practical limit you can work around.

CAN'T

Delete or merge anything

Why: A deleted job or a merged customer is hard to put back, so this is off the table for an assistant.

Instead: Delete or merge in Standard Time® yourself, with the confirmations it already asks for.

CAN'T

Edit time logs, expenses, or invoices

Why: These are your payroll and billing records, and changes to them should be deliberate and traceable.

Instead: Edit them in their own grids, with your normal rights. The assistant can still read them, if your rights allow.

CAN'T

Change existing project settings

Why: How a project is set up affects how its work is tracked and billed, so that stays in your hands.

Instead: Change them in the Properties panel. The assistant can create a new project or copy a whole one, if an administrator approves it.

CAN'T

See more than you can see

Why: It signs in as you, with your rights, and those are re-checked on every request. Projects you cannot see stay hidden.

Instead: If something is missing, ask an administrator whether you need another right.

CAN'T

See what was hidden when it connected

Why: Cost and rate fields and user names can be hidden at connection time, and a hidden field cannot be asked about.

Instead: To change what it sees, revoke the connection and reconnect with different choices.

CAN'T

Save a change without your approval

Why: Every change is a preview first, then a separate apply. A proposal that is not approved within 30 minutes lapses.

Instead: Read the preview before you approve, especially the dates and the number of rows.

CAN'T

Recall notifications already sent

Why: Undo removes what a change created and restores what it edited, but it cannot pull back a webhook or Zapier message that already went out.

Instead: If automations are listening, check your first changes on a throwaway project before you rely on them.

LIMIT

Make a very large change in one go

Why: Only one proposal is open at a time. A preview lists up to 25 rows, and one proposal creates up to 25 new tasks.

Instead: Work in rounds. Apply the first set, then attach the next tasks to the project it created.

LIMIT

Update a project that has opted out

Why: A project whose AI-updates setting is off is skipped.

Instead: Turn the setting on for that project, or make the change by hand.

LIMIT

Run on an on-premise install

Why: Claude Chat and ChatGPT connect to Standard Time® cloud accounts only.

Instead: On-premise users can still use the built-in AI Chat with the provider an administrator assigns.

Where It Can Still Be Wrong

The limits above protect your records, but they do not make an assistant infallible. It can misunderstand a request, pick the wrong job when two have similar names, or be confidently wrong about arithmetic. Your answers are also only as good as the data behind them, so a missing scan is still a missing scan.

  • Read the preview. It is there to be read, not skimmed.
  • Check the first few answers against your own grids, the way the Claude and ChatGPT smoke tests do.
  • Start read-only if you are not ready for changes.

The Two 30-Minute Windows

Approving and undoing are separate clocks. You have up to 30 minutes to approve a proposal, and once it is applied you have 30 minutes to undo it.

A timeline: preview shown with nothing saved, up to 30 minutes to approve, then 30 minutes to undo

After the second window closes, correct tasks by hand or ask for a new change.

What It Can Do

  • Answer questions about projects, tasks, and hours from your live data, within your rights.
  • Propose changes to project tasks: dates, planned hours, progress, assignments, links, and new tasks.
  • Show a preview, wait for your approval, and offer undo for 30 minutes.

See AI for Manufacturing for the full picture, or the AI Features FAQ for quick answers. Claude is made by Anthropic and ChatGPT by OpenAI; neither is affiliated with Standard Time®, and their plans and menus are their own.

← Back to Blog   Try Standard Time® Free

AI That Stays Inside the Lines

Start a free 30-day trial and connect an assistant with the limits you choose.

View Pricing Contact Us