# New hires spend weeks asking questions because we never wrote documentation. Can a bot answer them in Slack? URL: https://getunblocked.com/answers/can-a-bot-answer-new-hire-questions-in-slack-without-documentation/ Published: 2026-10-29T18:30:00Z Updated: 2026-10-27T23:52:59Z Author: Dennis Pilarinos Categories: Context Engine, Engineering Insights **Short answer:** Yes, if the bot reads where the answers already live: code, merged PRs, tickets, old Slack threads and incident channels. Slack's own AI answers from what was said in Slack. Wiki-first bots such as Notion AI and Guru answer from pages someone wrote and verified. A bot that reads the sources, cites the PR or thread, and keeps the asker's permissions can answer on day one. --- Yes, as long as the bot reads the places where the answers already are. The documentation was never written, but the decisions were made somewhere: in a pull request, a ticket, a Slack thread from the incident, a commit message. **A source-reading bot** is a Slack bot that answers from those systems directly, cites the PR or thread it used, and shows the asker only what they could already open themselves. **A wiki-first bot** is one that answers from pages a person wrote and verified. The question says you have no such pages. The choice is between a bot that waits for documentation and one that does not need it. ## Why new hires ask in Slack in the first place The 2026 Stack Overflow survey found developers look for specific answers in "the people you work with (72%), the code (63%), and documentation (60%)" ([Stack Overflow](https://stackoverflow.blog/2026/10/06/the-results-of-the-2026-developer-survey-are-here/), October 2026). Slack is where the people are. The write-up's own gloss on that number is that it "explains why senior devs spend half their day answering DMs." For a new hire the pattern is worse. They do not know who to ask, so they ask the channel, wait, and ask again next week when the answer was in a thread nobody linked. The missing wiki is not the cause. A wiki is a cache of answers to questions someone predicted. Even written pages drift: a 2026 study of 25,511 pull requests across 714 repositories opens by noting that README files "frequently become outdated" ([arXiv](https://arxiv.org/abs/2603.00489), 2026). The questions new hires ask are the ones nobody predicted: why the retry count is five, which of two auth services is current, who owns the deploy script. Those answers exist in history. The fix is a bot that reads it. ## What Slack's own AI answers from Slack's AI search answers are the closest thing to a built-in bot. Its guide says "Answers include citations that refer to the source messages or files that informed them," and "AI-generated responses only include information available to you" ([Slack help](https://slack.com/help/articles/25076892548883)). Slack lists search answers for Business+ and Enterprise+ plans. Enterprise search, on Enterprise+, adds connected sources such as Google Drive and GitHub. Slack's product page frames the scope: "if it's shared in or integrated into Slack, it's searchable" ([Slack](https://slack.com/features/ai)). The bot answers from what was said in Slack, plus whatever a connected app exposed. For example, if the retry count changed in a PR nobody discussed in a channel, the answer is not there. If three threads disagree, the search returns three threads. For a team whose decisions happen in Slack this covers a lot. For a team whose decisions happen in code review it covers the comments about the code, not the code. ## Wiki-first bots need the pages you did not write Notion AI, Guru and Glean can all answer in Slack. Each starts from a corpus someone curates. Notion describes matching your question against an index of your pages, and says "Notion AI can only ever use content you give it access to" ([Notion help](https://www.notion.com/help/notion-ai-security-practices)). Its enterprise search can also "Search across Slack, Google Drive, GitHub & more" and is marked beta ([Notion](https://www.notion.com/product/ai)). Guru's model is verification: answers are "Routed to expert for review" and content can be "Auto-verified as usage increases" ([Guru](https://www.getguru.com/)). The expert who verifies is the senior engineer the new hire would otherwise DM. Glean's bot answers @mentions in Slack and can reply unprompted where an admin allows it ([Glean docs](https://docs.glean.com/user-guide/embedded-integrations/slack/getting-started-user)). It reads Slack through a federated connector. Its docs say the connector "does not crawl or index Slack message content" and "always follows Slack's own access controls" ([Glean docs](https://docs.glean.com/connectors/native/slack/setup/slack-rts-connector), updated October 20, 2026). It also means Glean fetches Slack history live per query, not reasoned over with the code and PRs. All three work best once there is a verified page to point at. The question is how to answer during the months before anyone writes it. Our [AI documentation tools comparison](https://getunblocked.com/blog/best-ai-documentation-tools-engineering-teams/) covers the authoring side of that problem. ## What a good answer looks like, and who sees it A good answer does three things that search hits do not. - It cites the PR or thread, so the new hire can read the original. - It names who decided, so they know who to follow up with. - It says whether the decision still holds, because the thread from March may have been reversed by a PR in June. The answer on [what an agent believes when the wiki and the code disagree](https://getunblocked.com/answers/wiki-says-one-thing-code-does-another-which-does-ai-believe/) covers that last step. This is the shape Unblocked's Slack bot is built around. It indexes the channels an admin selects, with a backfill on connect and real-time updates after, alongside the repositories, tickets, docs and incident tools the team already uses. A new hire can @mention it in a channel, DM it, or let it auto-respond in channels where confidence is set high ([Unblocked docs](https://docs.getunblocked.com/data-sources/slack.md)). The answer lands in a thread with its sources, and teammates can correct it there. Olli Draese, Technical Architect at Cribl, puts it in one line. "Unblocked is our number one tool to find information we should know but don't." Permissions matter because the bot can read more than the asker. Slack's own AI holds to "information available to you," and Glean searches "private channels only for confirmed members of those channels." Unblocked treats public channels as public and limits private-channel content to members. A member must approve any answer built from private context before it posts in a public channel, and other bots or workflows get only public content. **Data Shield** is the Unblocked control that filters each answer to what the asking user may see; it is always on for Slack and cannot be turned off. Whatever bot you pick, test it with a new hire's account against a private channel they are not in. ## How to tell whether it is working Measure ramp time before you measure anything else. DX's April 2026 analysis (sample size 400 companies) puts the average time to a new hire's tenth merged PR at 33 days, down from 39 in Q4 2025 ([DX](https://newsletter.getdx.com/p/developer-ramp-up-time-continues), April 2026). Record it for the last three hires, then the next three with the bot in place. Our [developer onboarding tools comparison](https://getunblocked.com/blog/best-developer-onboarding-tools/) uses the same metric. Two smaller counts help in week one. Count questions in the help channels that get a bot answer and no human follow-up in the thread. Count the ones a human corrected, and read those, because that is where a source is missing or stale. If corrections cluster around one system, connect it. The [context layer guide](https://getunblocked.com/blog/build-context-layer-engineering/) has an order for which sources to connect first. The documentation can stay unwritten. Once the bot has answered the same question for the third new hire, that is a page worth writing, and you will have the thread to write it from. ## Read more - [8 Best Developer Onboarding Tools, Compared by Layer (2026)](https://getunblocked.com/blog/best-developer-onboarding-tools/) - [Best AI Documentation Tools for Engineering Teams (2026)](https://getunblocked.com/blog/best-ai-documentation-tools-engineering-teams/) ## Related answers - [There is drift between our repos and our documentation and agents cannot tell. What handles this?](https://getunblocked.com/answers/documentation-drift-between-repos-and-docs-for-ai-agents/) - [How do I mark old Confluence pages as deprecated so AI stops using them?](https://getunblocked.com/answers/mark-confluence-pages-deprecated-for-ai/) - [MCP tools are eating the whole context window in Claude Code. How do other teams give agents company context?](https://getunblocked.com/answers/mcp-tools-filling-claude-code-context-window/) - [We have context all over the place and constantly being generated. How do we organize it so AI pulls the right thing?](https://getunblocked.com/answers/organize-scattered-context-so-ai-pulls-the-right-thing/)