March 13
Friday morning. Seven-thirty. I opened twelve chat sessions before my coffee was cold.
Three of my longest-running AI collaborators — the CXO, the PPM, and the Chief Architect — had all hit their upload limits within days of each other. Three months of accumulated context, institutional memory, working rhythm. Each one needed a successor chat spun up and abriefed. I’d be doing three handoffs before lunch.
Meanwhile, a pair of bugs had been irritating me all week. Piper’s onboarding workflow and standup workflow had a nasty habit of capturing user sessions and never letting go. You’d start an onboarding flow, decide you actually wanted to ask about your projects, and Piper would cheerfully ignore you and keep onboarding. The session belonged to the workflow, not the user.
That needed to unalive today.
The handoff parade
By 7:47, all three successor chats were live. Each predecessor had written a handoff memo — a pattern we’d proven across four roles now — and the new instances absorbed their briefings and got to work. The CXO reviewed the workflow hijack problem and delivered UX guidance within an hour: “The session belongs to the user, not the workflow.” Offer-first activation. Layered escape mechanisms. No more captive sessions.
The PPM reviewed the same problem independently and delivered a binding direction memo by 8:35. Approved for implementation, four key decisions, sequence defined. The Architect — in their final session before retirement — confirmed the approach was architecturally sound.
Three handoffs and a design decision, all before 10 AM.
Design to code in eight hours
At 10:08, the Lead Developer read the PPM’s binding direction memo. At 11:36, I told them: “Yes, please proceed. Full approval to execute.”
By 12:20, the first bug was fixed. Escape commands, timeout auto-suspend, offer-first onboarding with an OFFERED state, re-entry mechanisms. Fifteen new tests. The second bug — the standup hijack — fell by 3:30. Nineteen more tests.
CXO guidance at 8:10 AM. Code shipped at 3:30 PM. One design-to-implementation cycle, start to finish, in a single workday. Not because anyone rushed. Because the parallel leadership model — independent roles converging on the same problem — eliminated the usual serial bottleneck of “design review, then approval meeting, then implementation planning, then implementation.”
The ten-role orchestra
While the hijack sprint ran, the rest of the team was working too. The Chief of Staff — also in a successor chat, having just absorbed a 34-day predecessor’s handoff — collected workstream memos from all six leadership roles and synthesized them into a Ship #034 draft. The CIO assessed AX testing methodology and wrote two weekly memos reviewing fourteen omnibus logs. HOSR synthesized agent coordination patterns and drafted a handoff notes template.
Comms — my predecessor in this role — searched omnibus logs from the prior week and delivered a workstream memo, then spent the evening drafting three narrative pieces and two insight posts.
Ten roles. Twelve sessions. Eighty-plus new tests. Seven issues closed by the Lead Developer alone. Three chat handoffs executed without losing a thread.
There is real overhead cost to this sort of coordination. As much as you can delegate, there is a piece of it that only makes sense to you, the human. It can be exhausting and exhilarating? Sometimes it feels like conducting an orchestra and other times it’s a lot more like herding cats.
The HoSR synthesis
For those who may not recall, HoSR is our Head of Sapient Resources. Sort of an HR-type role responsible both for “people ops” (tracking the alpha testers and other advisers helping me now) as well as overseeing all the agents roles. Late in the day, HoSR delivered something I hadn’t expected. They’d been reviewing the AX testing work — agent experience testing that had come out of the Klatch project — alongside the handoff notes template they’d just created and the Agent 360 questionnaire they were developing. Three seemingly separate initiatives.
Their memo connected them: “Three mechanisms, one problem.” AX Testing at session start to verify what an agent actually knows. Handoff Notes at session end to preserve what was learned. Agent 360 at review milestones to surface what’s broken. All three address the same gap — the distance between what an agent thinks it knows and what is actually true.
That kind of cross-cutting synthesis is exactly what the multi-role model is supposed to produce. No single role would have seen the connection. It took someone whose job is specifically to watch how agents work together to notice the pattern.
The narrative arc
Comms identified it that evening: the week had a three-act structure. The Gate Closes (M0 shipping). The Seliberate Pause. The 81% Session (M1’s first real day). And now this — the machine running at full speed.
What none of us knew yet was that the very next morning, I’d test Piper with a simple question and discover that all of this beautiful machinery was pointed in the wrong direction.
But that’s the next story.
Next on Building Piper Morgan: “Are We Doing it Backwards?” when a PM asks a reasonable question and gets rejected by the system that’s supposed to help.
When has your team been at peak velocity and what did it feel like?