Skip to main content

19 posts tagged with "threewill"

View All Tags

A Playbook for Client Success Manager Bots

· 2 min read
Mike Homol
Director of Innovation @ ThreeWill

Willy, my Grok Bot for the Operational Dashboard, has a way of working I trust: it plans first, works through cloud agents and PRs, and keeps the docs current. This week I wanted that same way of working for ThreeWill Managed Services clients, as Client Success Manager (CSM) bots, without starting from scratch each time.

What happened​

  • My Grok Bot Gongi and I turned the habits that work into two shared skills: CSM standards, the rules every CSM bot follows, and new CSM bot, the steps to set one up
  • The standards cover checking the client's current sprint before planning repo work, sending all repo work to cloud agents, plan first (except bug fixes), feature branches and PRs that I merge, keeping knowledge in git, and updating docs in the same PR
  • A new CSM bot starts by opening a plan-only docs PR for each of its repos, with the standard repo structure and a copy of the ThreeWill engineering standards
  • I stood up CSM bots for two Managed Services clients with the new skill, one on Nov 10 and one on Nov 15. The second client's repos had just been through cloud-prep
  • A new CSM rule now goes in one place, and every CSM bot picks it up the next time it loads the skill

cloud-prep: Getting Local Cursor Workspaces Ready for Cloud Agents

· 2 min read
Mike Homol
Director of Innovation @ ThreeWill

I want more of my repo work going to Cursor cloud agents. The catch: a cloud agent only sees the repo and its brief. The decisions behind the code mostly live in my local Cursor chats and plans, where no cloud agent can read them. So I built cloud-prep, a repeatable skill that moves that context into the repo once.

How it works​

  • Harvest: read the workspace's local Cursor chat history and plan files, read-only, and draft a context doc: purpose, architecture, key decisions with dates and reasons, conventions, known pitfalls, open issues, and environments (secret names only, never values)
  • Review: I read the draft and fix it before anything goes into the repo
  • Docs-only PR: a cloud agent checks the draft against the code and opens one PR. It adds the context doc, a decision log, the original plans, an AGENTS.md read-first list, and a copy of the ThreeWill engineering standards, including the plan-first rule. No source code changes
  • After that, the rule that docs get updated in the same PR keeps the context current

What happened​

  • My Grok Bot Gongi and I ran it on three repos this week: an internal HR app and two client solutions
  • One of the client solutions wasn't on GitHub yet, so it moved there with the context docs in its first commit
  • The review caught stale docs along the way, like wrong environment variable names in a runbook and an outdated package version, and I fixed them

Willy Ships a New Dashboard Nav and an EHR Connector

· 2 min read
Mike Homol
Director of Innovation @ ThreeWill

Willy, my Grok Bot for ThreeWill's Operational Dashboard, worked the same way all week: write a plan, wait for my approval, then ship. Two big ones landed on Nov 14. The credit for the work goes to Willy. I picked the direction, approved the plans, and merged.

The Operational Dashboard's new Concept B navigation

An early look from the dev site with sandbox data, taken before the final polish. A more polished version is coming in a future post.

A new navigation: Concept B​

  • A UX teammate groomed a set of navigation stories on Nov 12
  • Willy turned them into a plan with three mockup concepts, and I chose Concept B: capsule tabs across the top plus a slim sub-bar
  • The left rail is gone. There are four tabs now: Dashboard, Communities, Scorecard and Settings
  • A category filter on the Dashboard replaces the seven category links that used to sit in the nav
  • Willy built it as 2.3.6.20, then did a polish pass: flatter tabs, a more compact Scorecard, and fewer extra links. 2.3.6.20 is in testing now

An EHR connector over FHIR​

  • The dashboard's Sync service needed its first connection to an EHR for a demo, using an EHR vendor's FHIR API
  • Willy wrote the plan, I approved it, and Willy shipped the connector as Sync 0.4.0
  • Testing against the vendor's sandbox caught a bug right away: Sync only read the first page of patients, because the next-page link in the vendor's FHIR R4 responses is labeled differently than the code expected
  • Sync 0.4.1 fixed the paging the same day
  • The connector feeds four demo measures: Active Clients, Services Delivered, Clients Seen and Engagement Rate

Willy's Two-Day Dashboard Sprint

· One min read
Mike Homol
Director of Innovation @ ThreeWill

Willy, my Grok Bot for ThreeWill's Operational Dashboard, had a fast week. Over Nov 3 and 4 it planned and shipped a batch of changes, and each one started as a plan I approved before any code was written. The credit goes to Willy. I reviewed and merged.

What happened​

  • Executive Overview cadence tabs: the overview tiles now follow the Scorecard's cadence tabs, and empty tabs are hidden
  • A simpler settings pane: instead of fixing a filter override that never applied, Willy proposed removing those settings altogether, and I agreed
  • Fixed built-in navigation: the main pages can't be accidentally hidden or reordered anymore, and extra items are added as links
  • Throttling fixes (2.3.6.10): retries that respect the server's back-off, plus a queue for check-in saves, to cut down on SharePoint throttling. Saving the Scorecard went from about 50 server requests to 1 or 2
  • A refetch fix (2.3.6.11) for the Scorecard after a check-in save. It passed the QA regression run
  • Willy also refreshed the roadmap to match. That was three PRs in two days, and I merged all three

AI on Client Delivery

· One min read
Mike Homol
Director of Innovation @ ThreeWill

Most of my week at ThreeWill was client delivery, and Cursor was in the middle of almost all of it. Here are a few of the ways AI helped, with the clients kept anonymous.

What happened​

  • Access rules: I wrote up how access rules work in a records app, with a diagram, so the client could confirm we understood them
  • Azure Maps: I set up the plumbing so the app's UI can render maps
  • DevOps cutover: I wrote an Azure DevOps cutover walkthrough to move a client's pipelines into their own organization

Managed Services Goes Under Source Control

· One min read
Mike Homol
Director of Innovation @ ThreeWill

ThreeWill's Managed Services work runs on a Power Platform solution. Until this week, it didn't live in source control. Now it does.

The Monday check-in page

What happened​

  • I set up a Managed Services repo and exported the ManagedServicesCore solution into git, with each new version saved as it ships
  • I added a Monday check-in custom page to the solution, with tabs and choice fields for the weekly review
  • Working in Cursor with the solution unpacked in the repo makes each change easy to see and review

Willy Keeps the Operational Dashboard Moving

· One min read
Mike Homol
Director of Innovation @ ThreeWill

Last week I introduced Willy, the Grok Bot that looks after the codebase for ThreeWill's Operational Dashboard. This week the credit goes to Willy. I reviewed and merged.

What happened​

  • Willy wrote up the team's sync architecture decisions as a set of decision records, including how sync conflicts get surfaced
  • Willy split the repo into a cleaner monorepo with path-filtered CI, so a change only builds the parts it touches
  • Willy documented the environments and wrote feature and regression-test docs for the latest hotfixes, plus onboarding notes for UX contributors
  • Every change came in as a pull request: Willy opened 5 PRs this week (Oct 26 to Nov 1), and I reviewed and merged all five

Bots Running the Work Week

· 2 min read
Mike Homol
Director of Innovation @ ThreeWill

A couple of weeks ago I started leaning hard on Grok Bot. This week a couple of those bots started carrying real weight in my ThreeWill work week — not by doing the work for me, but by keeping the busywork moving so I can spend my time on the parts that need me.

The pattern is the same for every bot: it drafts, I approve, and only then does anything get posted, filed or merged.

Willy, the roadmap bot​

Willy maintains the codebase for ThreeWill's Operational Dashboard.

  • Keeps the roadmap and backlog in sync with what the team decides, including a roadmap PDF with a requirement ID on every row and a "what changed" section, and waits for my review before anything goes to SharePoint
  • Opens docs-only pull requests through a cloud agent so the repo's roadmap and backlog match what we published, with app code and store-frozen paths off limits; I review and merge
  • Posts roadmap updates to the team, clearly labeled as Willy, only after I approve the wording
  • Even trades notes with a teammate's own Grok Bot, which was a fun first

Rudy, the operations bot​

Rudy handles the operational side of my week.

  • Daily timesheets: every weekday morning Rudy drafts yesterday's time from my calendar, Cursor sessions, chats and work items, and builds a tentative plan for today; I adjust, and Rudy posts it after I OK it
  • Retention scoring: we replaced a hand-run AI prompt with a transparent, documented formula for how likely each Managed Services client is to stay, and set up a monthly grooming routine that recommends changes, waits for my OK, writes them back, and updates a "clients at retention risk" scorecard measure
  • A running to-do board with check-ins during the day, so nothing slips between meetings

Why it works​

None of this is magic. It's small, repeatable routines with a human gate at the end. The bots are great at gathering evidence and drafting; I'm still the one who decides.

ThreeWill Feud Goes Open Source

· 2 min read
Mike Homol
Director of Innovation @ ThreeWill

Last week the Feud board ran live at ThreeWill's corporate QAD. This week I pulled it out into its own product: threewill/feud is now a public, MIT-licensed repo that anyone can use to run a Family Feud–style show for their own team.

The big change is that the board isn't tied to one event anymore. Questions load from a pack — one JSON file — so the next show is just a new pack.

What happened​

  • Productized Feud as pack-driven boards: title, family names, live and backup boards, win rules and optional lightning questions all live in one JSON file
  • Added a Microsoft Forms importer that clusters similar survey answers into a pack, with a review file of the raw strings so you can tidy labels before showtime
  • Let the host set the win score and turn the lightning round on or off, and split showdown X's from the family's three strikes
  • Added an MIT license, a README that walks through gameplay, and host and audience screenshots
  • Added Feud to Creations on homol.work

Host view — answers, strikes and round award stay on the operator's screen

Audience view — the shared board with numbered slots, scores and the clock

Try it​

Clone threewill/feud, run npx serve ., open the host page, then pop out the audience board and share that window. It's not official Family Feud — no licensed logos, and the ding and buzzer are synthesized stand-ins.

ThreeWill Feud — Live Board for QAD

· 2 min read
Mike Homol
Principal Consultant @ ThreeWill

For ThreeWill’s corporate Feud at QAD I needed more than a slide deck: a live board the room can watch, scoring that feels like Family Feud, and a host surface I can drive without stealing the share. This week that board ran live — one tab shared with everyone in the meeting, another tab on my machine that updates the shared experience.

The QAD rounds came from survey results we imported for that event. The board itself is not tied to those questions: any answer bank can be loaded, so the same host + audience split is reusable for the next Feud (or anything that wants a two-window live board).

What happened​

  • Built a live Feud host + audience board (HTML/JS/CSS) with ding/buzzer sounds and board state for multi-select answers
  • Ran it live at October QAD: share the audience tab with the meeting; keep the host tab private so reveals, strikes, and scoring stay on the operator
  • Rescaled multi-select boards to a 100-point pie and added an audience clock so pacing stays visible
  • Kept the audience window painting correctly in Edge and made hopping between host and board reliable
  • Documented the Feud End game host button and a clustered QAD runbook so setup is repeatable

Host start — teams, backups, Open audience board

Audience start — shared contestant board before a round

Host question — answers and survey source visible only to the operator

Audience question — the room sees the prompt, numbered slots, and clock