AI MissionsCase StudyProductivity

Project Knowledge Hub in Practice: A Tuesday, Rewritten

HE
Harry Edwards · Head of Solutions Engineering
January 11, 2026

I run solutions engineering, which means I spend my days in other companies' actual workflows — not the tidy version on the slide, the real one with the 7:40am scramble before a steering committee. So let me not pitch you an abstraction. Let me walk you through one Tuesday with a real customer, a mid-market fintech mid-way through a payments-platform migration, and what changed when we put the Project Knowledge Hub in front of their program lead. I'll call her Dana, because that's her name.

The Tuesday before

Dana's steering committee meets at 9:00am. The one question the sponsors always open with is the one that sounds trivial: is the migration on track, and if not, why? Before the Hub, answering it was a ritual.

She started at 7:35. Jira first — the migration epic showed 68% of tickets closed, but Dana knew two of the "closed" ones had been reopened in a thread she'd half-remembered, so that number was soft. She switched to git to see what had actually merged over the weekend; three PRs, one of them a revert, which meant a feature the board still counted as done was, in fact, backed out. She pinged the tech lead on Slack to confirm — no reply, he was heads-down. She dug through email for the payments vendor's note about the sandbox certification date, because a slip there was the real risk and it lived nowhere near Jira. She found a Confluence page describing the cutover plan and couldn't tell if it was current.

By 8:50 she had a paragraph. It was 70% right and she knew it. She walked into the committee and led with a number she didn't fully trust, and when a sponsor asked "why is the certification date at risk," she had to say she'd follow up. That follow-up cost another two people an hour each that afternoon. This is the part that never makes the ROI slide: the escapes. The blocker Dana didn't surface on Tuesday because it was buried in email became the reason the cutover slipped a week. Not because nobody knew — because the knowing was scattered.

The same Tuesday, with the Hub

Now the same morning. At 7:38 Dana opens the portal and types the question she'd otherwise spend seventy-five minutes reconstructing: "Where does the payments migration stand, and what's the top risk to the cutover date?"

She watches the observations stream. The Mission routes to the Jira Agent — open blockers, reopened tickets flagged. Then the Git Agent — it catches the weekend revert and notes that the "done" feature is actually backed out, reconciling the board against what shipped. The Confluence Agent pulls the cutover plan and stamps its last-edited date so Dana knows how fresh it is. The Comms Agent surfaces the vendor email about the sandbox certification slipping two days, and the Slack thread where the tech lead flagged it Friday afternoon. Ninety seconds later she has a grounded, cited status: 68% closed but one counted feature reverted, real completion nearer 64%, and the top risk is the certification date, with the vendor's own email linked as the source.

She walks into the 9:00 committee with an answer she trusts, sourced line by line, and when the sponsor asks "why is the date at risk," the email is already right there in the status. No follow-up. No two people losing an afternoon. The blocker that would have escaped got surfaced four days before it could bite.

Before — ~75 min, ends unsure Jirasoft % gitrevert? Slackno reply email digvendor? Confluencestale? 70% sure After — ~90 sec, grounded & cited 1 question Jira Agent Git Agent Confluence Comms Agent trusted Outcome: revert caught, vendor slip surfaced 4 days early ~75 min → 90 sec per status · zero follow-up afternoons · one cutover-slip escape avoided

What it actually saved — the honest numbers

I'm allergic to ROI math that only works on a slide, so here's what we measured on this account over the first six weeks, not what we hoped.

  • Per-status time: ~75 minutes down to ~90 seconds. Dana ran the Hub roughly a dozen times a week — before every stand-up, every sponsor ping, every "quick question" from a peer lead. That's around 14 hours a week of her time returned, and she's the person whose judgment the company is actually paying for.
  • Follow-up afternoons: gone. The Tuesday ritual used to spawn one or two "let me get back to you" threads a week, each burning an hour from two other people. Because the answer arrives already cited, the follow-ups mostly stopped — call it another 3–4 hours a week across the team.
  • One escape avoided, and that's the big one. The certification slip that the Hub surfaced four days early was, in the prior quarter's equivalent, exactly the kind of thing that cost a full week of cutover schedule. One avoided week of slip on a migration with a room full of engineers waiting on it dwarfs every hour of status time combined.

The pattern generalizes. Every team we've rolled this out to has the same shape of problem — the truth exists, it's just scattered, and a senior person is the integration layer. Give them a Mission that does the finding, grounded in their own enterprise knowledge and wired to their own tools through our enterprise integrations, and the status question stops being a tax.

Getting it live

The part solutions engineers love: standing this up is a configuration exercise, not a build. We register an instant MCP server per source system — Jira, Confluence, git, Slack, email — point a source-specialist agent at each, and the Mission is finding within a day. Because it's read-only, there's no approval gate to design and no risk of the Hub touching production data; it queries and reports, nothing more. Teams that later want it to act — post a status back, notify a stakeholder — flip on the human-in-the-loop path, and those actions route through the decision queue for a human's yes. But most start, and stay, in pure finding, because that alone gives them their week back.

If you want the leadership case for why this matters, Ajay wrote why it matters. And if you want the engineering detail behind the two-tier reasoning, Trevor's how it works is the one to read next.

Discussion

No comments yet — start the conversation.

Join the discussion

See StudioX run.

Put autonomous AI workers to work on your own systems and knowledge.