Your Program Managers Aren't Program Managers. They're Meeting-Minute Factories.
Your senior program managers spend most of the week manufacturing artifacts about the work, not landing engagements. Here is the operating layer we built, and the boundary we hold.
Published June 17, 2026 by David Suydam, Abi Fagbuargo
Walk through your delivery org and look at what your senior program managers actually do with their week. You hired them to land engagements on time and on scope. They spend most of the week manufacturing artifacts about the work.
Run the check on your own delivery:
Your senior program manager is collating status updates by hand the night before they go to the client.
They're rewriting minutes the notetaker got 70% of the way right.
They're spending two days scaffolding a Confluence kickoff when the SOW already says what goes in it.
They're scrambling at 9am to figure out what's blocked before standup.
They're chasing a RAID log no one updates between meetings.
What actually moves the engagement forward is relationship management, problem-solving, change management, coaching the team. That's the work that doesn't fit. PMI's 2024 research puts a number on the gap: 61% of senior leaders say their teams need more of exactly that layer; organizations allocate only 25% of training hours to it. The admin layer is where program manager time goes. The relationship layer is where program manager value lives. Closing that gap is the workflow we redesigned for ourselves.
The workflow we rebuilt
One Architech Senior Program Manager, Abiodun Fagbuaro, piloted a delivery operating layer on a real two-week client engagement. Statement of work in, structured Confluence and Jira out, in minutes. Transcripts in, draft minutes and a live RAID log out. Jira state plus RAID state in, client-ready status report out. Every artifact reviewed by the program manager before it goes to the client.
Two weeks. Four deliverables. Nine standups. Nothing dropped.
What this replaces is the time the program manager used to spend manufacturing artifacts. What it frees up is the work most of us became program managers to do.
How it actually works
The operating layer has three concrete pieces, sitting on top of Anthropic's public Agent Skills pattern:
A foundation file. A CLAUDE.md-style workspace foundation defining client conventions, naming rules, formatting standards, and each program manager's writing nuances, so drafts arrive in the program manager's voice rather than a generic AI voice.
A navigation file plus shared templates. The routing layer that maps an intent ("update the RAID log") to the right skill and the right Architech template. Consistency across program managers and clients.
A skills library. 20+ skills, with around 10 universal across the program manager team and the rest engagement-specific. Each skill is its own SKILL.md folder with instructions, conventions, and procedure.
On top of that scaffold, the capabilities the program manager actually uses day to day:
Drop-the-SOW kickoff. Statement of work in; structured Confluence space and Jira project out in minutes, matching the documentation structure our practice areas agreed on two years ago.
Transcript intake with three sub-agents. One files the transcript. One drafts the minutes. One updates the live RAID log, closing risks that resolved, recording decisions that landed, surfacing new items.
Status reports that assemble from Jira plus RAID. The program manager reviews and edits. The program manager no longer authors from scratch.
A 9am daily briefing that scans active projects, surfaces blocked tickets and top risks, and lands before standup.
A project-plan generator with three thinking partners (product owner, UX, tech lead) for clients who want a traditional plan alongside the agile work.
A scope-creep agent that watches conversations and deliverables for drift from the SOW and helps frame the change-request conversation when drift is real.
Where we draw the boundary
Jira and Confluence are connected via API. Microsoft 365 and Outlook are deliberately not. A human reviews every artifact before it goes to the client. Regulated buyers screen on this combination.
The threats are real and current. Microsoft 365 Copilot had a zero-click vulnerability (EchoLeak) that let a single email exfiltrate chat history. AI notetakers face class actions over recording consent and biometrics (In re Otter.AI and parallel Fireflies BIPA suits). On meeting minutes specifically, corporate-governance lawyer Douglas Raymond named the failure mode last August: AI "lacks the ability to interpret context, tone and nuance," and "the presence of the tool may naturally lead to decreased vigilance" (Directors & Boards, August 2025). Human review is the control.
"This doesn't replace program managers. It moves us up, to the work most of us became program managers to do: relationship management, problem-solving, change management, coaching the team." Abiodun Fagbuaro, Architech Senior Program Manager
The proof gap in our category
AI-services firms do not generally publish first-person, named-workflow, mechanism-level redesigns of their own delivery, with the human-review discipline shown and the boundaries named. This piece is one. It is the third in a row from us, after the content-pipeline Customer Zero piece and the one-claim test. When operators are evaluating AI-services firms, "do you run this on your own delivery?" is becoming the first filter.
"What Abi did with this operating system is what 10x looks like inside Architech now: a program manager who found his superpower and used it to free himself for the work that actually moves clients. That's the bar. It's what I expect, what our clients are starting to demand, and what makes us worth hiring over anyone else." David Suydam, CEO
What we would do differently
One program manager piloted this on one engagement. Rollout to the broader team is the next phase, soon to be complete. The package (foundation file, skills library, navigation, templates) is built; the harder work is putting it in the hands of every program manager in the shape each of their engagements needs. Pilot-to-production is where most of the category stalls.
The M365 boundary stays. Measured outcomes: two weeks, four deliverables, nine standups, nothing dropped.
Why this matters when you hire an AI-services firm
This is what an outcome-driven Workflow Transformation Sprint looks like when we run it on ourselves. Outcomes are how we know it worked: four deliverables shipped, nine standups run, nothing dropped on a two-week engagement. Admin layer absorbed. Relationship layer freed. The connections that matter made; the connections that carry risk deliberately not.
When you evaluate an AI-services firm next, ask two questions. Does the firm run this on their own delivery? When the AI drafts the minutes, who reads them before they go to the client? Here's how we run it for you.
Ready to apply this to your workflows?
Architech's AI Jumpstart is the structured entry point.