To bot or not to bot: what do you want it to do after you ask?
Five documented business products, one question. Whether a bot belongs in your company depends on the behavior the job needs: an answer now, a job taken away and brought back, or agreed work that keeps running.
Key takeaways
- Deciding whether a business needs a bot starts with the behavior the job needs: an answer in the same conversation, a task handed over and returned, or agreed work that keeps running on a schedule or an event.
- As of September 5, 2026, xAI Grok Bot, Anthropic Claude Tag in Slack, OpenAI workspace agents, Google Workspace Studio, and Microsoft Copilot Studio event triggers each cover one or more of those three behaviors on business plans companies already pay for.
- Autonomy has an identity cost: Microsoft Learn's event triggers documentation, last updated February 19, 2026, states a Copilot Studio agent running on an event uses the agent maker's credentials, so the maker account should be treated as a service identity.
- Governance controls are often tier-dependent rather than product-wide, so the audit logs, allowlists, and directory sync a company assumes are present may sit only on the enterprise tier.
- Nova3's operational read is to pick one job that runs at least weekly, write down how it is done today, define what finished means, and check the product's results directly rather than trusting its own summary.
Review
We get asked some version of "should we have a bot?" most weeks. The useful question underneath it is: what do you want it to do after you ask? Three behaviors cover nearly every request. Help me now: you ask in chat and get an answer or a draft in the same conversation. Take a job and bring it back: you hand over a task, the product works on its own for a while, and it returns with a result or a question. Keep doing agreed work: the product runs a routine on a schedule or when an event fires, without anyone asking each time. These overlap, and none of them is a maturity level; they are different jobs. The five products below each cover one or more of these behaviors as supplied, on plans that many businesses already pay for. Four honest outcomes follow from matching a real job to a behavior: use the product as supplied, configure it, connect it to something it cannot reach on its own, or do not adopt it for that job. Nova3 configures, connects, and builds bots of all three kinds for clients, and this is the framing we use when the question comes up. Our read: pick one job that runs at least weekly, write down how it gets done today, define what finished means, and run the product against that standard for a short period with the results checked directly, not from the bot's own summary. A custom build is one of the four outcomes, not the default.
Dated events and sources
xAI made Grok Bot available for enterprises
Fact: xAI's September 3, 2026 announcement says Grok Bot is now available for enterprises. It describes each Bot as a worker created for a specific job that runs on its own computer in the cloud, is messaged like a coworker, and comes back when the work is done or it needs a decision. The release adds access, network, and audit controls, and gives Grok and Cursor Enterprise customers free usage for two weeks.
Nova3 operational read: This is "take a job and bring it back" and "keep doing agreed work" sold in one product. Before adopting, settle which tier you are on: Cursor's Grok Bot security documentation lists audit logs, the MCP allowlist, and SCIM as Enterprise-only, and says self-serve Teams does not see those settings. That is a configuration and suitability question, not a reason to dismiss the product.
Uncertainty: Customer names, adoption counts, and use cases in the announcement are vendor claims. The free usage window is time-limited.
xAI: Grok Bot for EnterpriseVendor source headline: xAI made Grok Bot available for enterprises
AI agents, permissions and ownership · Scope the controls for an AI workflow
Also cited: Cursor docs: Grok Bot security
Anthropic introduced Claude Tag for Slack
Fact: Anthropic's June 23, 2026 announcement introduces Claude Tag: Claude joins Slack as a team member, anyone in a permitted channel can tag @Claude and delegate a task, and the result comes back in the thread. It is available in beta for Claude Enterprise and Team customers. Anthropic's routines documentation adds scheduled jobs, channel watching, and subscriptions to individual pull requests.
Nova3 operational read: Tagging covers "help me now" and "take a job and bring it back" in one place; the routines turn it into "keep doing agreed work" for specific event types. Record the exact events you rely on (a schedule, a watched channel, a pull request) rather than treating it as general-purpose automation.
Uncertainty: Beta on two plans as of the source date. The claude.com product page lists Microsoft Teams as coming soon, not available. Anthropic's internal usage figures are its own.
Anthropic: Introducing Claude TagVendor source headline: Anthropic introduced Claude Tag for Slack
AI agents and recurring work · Bring Nova3 a recurring workflow
Also cited: Claude docs: Claude Tag routines and proactivity; Claude: Claude Tag product page
OpenAI documents ChatGPT workspace agents in research preview
Fact: OpenAI's workspace agents walkthrough, checked September 5, 2026, describes them as shared agents that can run repeatable workflows, use connected apps, follow skills, run on a schedule, and be shared with colleagues in the workspace. It says they are in research preview for ChatGPT Business, Enterprise, and Edu customers. OpenAI's scheduled tasks documentation says that on eligible plans a task can also run when a supported Gmail, Slack, or GitHub event occurs, and that Slack reactions, edits, deletes, and direct messages are not supported triggers.
Nova3 operational read: A shared agent on a schedule is "keep doing agreed work" without custom software. Because it is shared, decide who owns its connected apps and what they can reach before rolling it out to a team.
Uncertainty: The walkthrough is undated and labeled research preview. The scheduled tasks page says "eligible plans" for event triggers without naming them. Confirm access in the actual workspace before planning around either.
OpenAI: Building workspace agents in ChatGPT to complete repeatable, end-to-end workVendor source headline: OpenAI documents ChatGPT workspace agents in research preview
AI agents and recurring work · Bring Nova3 a recurring workflow
Also cited: OpenAI docs: Scheduled tasks and event triggers
Google lists the Workspace editions eligible for Workspace Studio
Fact: Google's Workspace Studio help page, checked September 5, 2026, says a work account needs an eligible Google Workspace edition or a Google AI expansion add-on, and names Business Starter, Standard, and Plus; Enterprise Standard and Plus; and Education Fundamentals, Standard, and Plus. Google's Ask Gemini in Google Chat help page says a skill must be created and turned on in Workspace Studio before it can be used in Chat, and labels some features as available through Gemini Alpha for work accounts enabled by the organization.
Nova3 operational read: For a company already on Google Workspace, configured flows in Workspace Studio are the "keep doing agreed work" route inside the existing subscription, and Ask Gemini in Chat is "help me now". Check the edition and the admin settings before assuming either is switched on.
Uncertainty: Help pages are undated and change without notice. Alpha and Experiments labels mean the feature set can differ by tenant and account type.
Google: Get started with Google Workspace StudioVendor source headline: Google lists the Workspace editions eligible for Workspace Studio
Model Context Protocol and connected tools · Evaluate an AI implementation partner
Also cited: Google: Ask Gemini in Google Chat
Microsoft documents autonomous event triggers in Copilot Studio
Fact: Microsoft Learn's event triggers overview, last updated February 19, 2026, says event triggers let a Copilot Studio agent act autonomously in response to a defined event, unlike topic triggers that require input from a user, and that event triggers can only use the agent maker's credentials for authentication. Microsoft's Teams support page, updated February 26, 2026, says Copilot in Teams group chats is part of the public preview program and that anyone with a Copilot license can chat with Copilot one-on-one.
Nova3 operational read: Autonomous runs act as the maker, not the person who asked. Anyone standing up a Copilot Studio agent for a team should treat the maker account as a service identity and review what it can reach. Copilot in Teams chats is "help me now"; Copilot Studio is the separate configured route for "keep doing agreed work".
Uncertainty: The dates are Microsoft's page update dates, not feature launch dates. Preview labels, licensing, and the billing impact Microsoft notes for event triggers can change.
Microsoft Learn: Event triggers overview, Copilot StudioVendor source headline: Microsoft documents autonomous event triggers in Copilot Studio
AI operations, explained · What Nova3 builds
Also cited: Microsoft Support: Microsoft 365 Copilot in Teams group chats
Questions this edition answers
Does my business need an AI bot?
Ask what you want it to do after you ask. If you need an answer or a draft inside a conversation, a chat assistant on a plan you already pay for is usually enough. If you need a task handed over and returned, or a routine that keeps running without anyone asking, you need a product that supports that behavior or a build that does. A custom build is one of four outcomes, not the default.
What is the difference between an AI assistant and an AI agent?
An assistant answers inside the conversation you are already having. An agent takes a job away, works on its own for a while, and comes back with a result or a question. A third pattern keeps doing agreed work on a schedule or when an event fires, with nobody asking each time. These overlap and none of them is a maturity level; they are different jobs.
Which business AI products can run work on their own as of September 2026?
Documented as of September 5, 2026: xAI Grok Bot for enterprises, Anthropic Claude Tag with routines in Slack, OpenAI workspace agents in research preview for ChatGPT Business and Enterprise, Google Workspace Studio on eligible Workspace editions, and Microsoft Copilot Studio event triggers. Availability, plan names, and preview labels change often, so confirm them in your own account.
What should we check before turning on an autonomous agent?
Confirm whose identity the agent runs as, because autonomous runs commonly act as the maker rather than the person who asked. Confirm which governance controls, such as audit logs and connector allowlists, exist on your tier rather than the vendor's top tier. Then run it against one real job for a short period and check the output directly.
When is a custom build the right answer instead of a supplied product?
When the job needs something the supplied product cannot reach or cannot be configured to do, and the job runs often enough to be worth the build. The four honest outcomes are: use the product as supplied, configure it, connect it to something it cannot reach on its own, or do not adopt it for that job. Building is the last resort, not the starting point.
Editorial method
Current State is a weekly review. Nothing appears here without a first-party source and an explicit event date.
/current-state.md mirrors the latest edition.