The ten-person check for @ChatGPT in Slack

You can add @ChatGPT to a Slack channel in one mention. The hard part is deciding which channels may call it, which licence pays for it, and which usage tier survives the next limit change.

For a ten-person startup, the answer should fit in one page that survives a limit change, a new hire, and a meeting where someone asks why the bot is in the sales channel.

Set the channel boundary before the first mention

Start with where the answer can appear. Who types matters less than where the answer lands, because a shared assistant is most dangerous when it can be summoned in a place the team did not choose.

OpenAI enabled Slack and Microsoft Teams users to invoke @ChatGPT in channels, threads, and direct messages, with no individual ChatGPT licence required for each teammate. That breadth makes the channel list the first field in the decision log.

Business and Enterprise customers can build ChatGPT teams and Team Tasks that execute on a schedule or after an event, such as a new Slack channel message. If the team wants that behavior, the channel decision has to include the trigger, not just the destination.

Private chats are the hardest to audit because they leave no shared record. Threads are easier to contain than whole channels. Set channel permissions with a short allowlist: a general channel, a project channel, and no private chats unless a named owner can review them.

Name the licence route that pays for the seats

The bill sits in the licence route. A shared tool can look free in a channel while the cost is carried by a plan nobody reviewed.

Re-count the seats that need access, then choose between individual licences and a shared team licence. The log should say which route pays for it and name the owner who can answer who pays, who can revoke, and who sees the usage.

If the answer is the founder's card, the licence route is not operational. ChatGPT Space is replacing Library as the shared location for team files and documents. The licence route should also cover where the team puts the files the assistant is allowed to read.

Choose the usage tier before the limit change

The usage tier changes under you. The next cut can turn a plan that is affordable this quarter into a different plan next quarter.

Before the change, write down what the team actually uses. If the current tier is close to the old limit, the cut becomes a budget event. If the team is far below the old limit, the cut may wait until the next hiring round.

Beginning October 30, 2026, the $200 Pro plan's bundled ChatGPT Work and Codex usage will be cut from 20 times the Plus allowance to 10 times, and weekly GPT-6 Pro chat messages will fall from 200 to 100. Current Pro 200 subscribers retain their existing limits through October 29, 2026, and receive a single $2,500 usage credit that expires at the end of 2026.

OpenAI launched a $500 monthly Pro 500 tier, the sole Pro tier with Ultrafast, which the company says can produce up to 300 Codex tokens per second, up to eight times faster than standard. It includes OpenAI's highest usage limits and Ultrafast mode, which allows GPT-6 Astra to provide faster responses in ChatGPT Work and Codex. Log whether Pro 200 or Pro 500 covers the team before that date.

Keep the three-field decision log

Keep the log in a document the team can open in a meeting. It needs a boundary, a payer, and a review date. If a field is blank, the decision is not done.

• Allowed channels: name the chat spaces where the mention is permitted, and treat every other space as blocked.
• Licence route: record the licence route that pays for the shared access, and name the owner who can revoke it.
• Usage tier: set the current tier, the limit-change date, and the review date before the next cut.

Put the log in the same place as the onboarding notes and the expense policy. A new hire should find it without asking. A departing employee should not be able to leave the tool running in a channel nobody remembers.

Review the log when a new channel is created, when a licence renews, and when a limit changes.