Figure Out What I Think
I suck at making decisions. Not all decisions... I'm pretty good with easy ones, the ones where there's a clear answer or at least a clear direction. Those I'll rip through. But the ones that actually matter... the taste calls, the judgment calls, the ones where the answer depends on something I haven't articulated yet... those I have to sit with. I'm an introvert. I process.
It takes a while.
AI and codegen have made me wildly productive. I don't write code anymore - I write specs and manage a team of (usually Claude Code) agents who build for me, four to twelve projects running on a given day. The speed is incredible. Things that used to take weeks take hours.
But speed isn't velocity. Velocity is speed plus direction. And when direction is clear, the velocity is massive... brainstorm, spec, agents build, ship. Harper laid out the spec-driven workflow almost a year ago and it's become how a lot of us work. Obra's brainstorming skill does a socratic interview to help build a spec that is legitimately the biggest unlock to my productivity in the last decade. When I know where I'm going, the tooling gets me there fast.
But I sit with decisions.
So all that speed... just waits for me. Or worse, it doesn't wait, and I pick a direction because the tooling is asking me to pick a direction, and I build the wrong thing fast.
Sitting with decisions
"Sitting with decisions" means something specific for me. I'm a Quaker. I've sat in meeting for worship most of my life... Friends schools, living and working at Pendle Hill, clerking the board of a Quaker school, going to Meeting most weeks. In worship you often have a thing you're working through. So do the people around you. Rarely is it the same thing. But something about being in that space, hearing other people speak out of silence about what they're carrying... it's often inspiring to look at your own thing differently. To let it season.
I realized a while ago that much of my decision-making, my leadership process, basically everything about how I navigate hard calls is deeply rooted in being a Quaker. Quaker business process is relatively well-defined (decisions made without voting, a clerk discerns unity instead of counting opinions, unresolved things get laid aside to season rather than forced through).[^1] And it reflects how I naturally make decisions even when I'm not in a meetinghouse. Leave space. Listen for what's underneath the obvious positions. Don't force convergence before it's ready.
The gap
So here's the thing. All of this tooling... the brainstorming, the spec-driven workflows, the agentic loops... it optimizes for speed. And speed is great. But I don't always work that way. I can't always work that way. The brainstorming gives me four choices and asks me to pick. And the decisions that actually need me (the uniquely human, nuanced, under-defined, overwrought me) don't fit in a multiple choice menu.
I found things wanting.
Inspired by brainstorming (which I love, genuinely) and by my own interpretation of Quaker process, I built a Claude Code skill. It is not optimized for speed. It is optimized for how I decide and how I make sure those decisions are good.
In that way, it absolutely does optimize for velocity. Moving quickly in the wrong direction is slow.
How it works
I form a committee.
Well, Claude and I do. We bring together voices with a stake in the question. Claude proposes some, and I tend to add more. The people I add seem to give voice to nagging concerns I have that I'm still processing... tensions I can feel but can't name yet. I may not be able to say "I'm worried about the first-moment framing problem." But I am able say "we need someone who feels what it's like to arrive cold."
Then I let Claude clerk the meeting.
That means I shut the fuck up and listen.
Each voice speaks once. I don't respond. I don't argue. I don't refine. The clerk checks in after each one ("anything arising, or shall we continue?") I note how I react, the things that the speaker's sharing brings up for me, and I and hold onto those. Sometimes I share those as an aside to the Clerk - more of a note to myself, sometimes I think and process, mostly I say "continue" because the discipline of holding those thoughts and feelings and NOT rushing to reaction is the point. When all voices have spoken, I get my turn. Then the clerk listens for where there's genuine unity and names what's unresolved.
This likely makes no sense unless you see it.
A session
I was building a college exploration app (kind of a "map the whole student" and find "hidden" paths). Deep into the build, I needed to rethink the chat interface. I had opinions: a persistent chat widget, conversations scoped to the page, rework the onboarding. It wasn't a bad plan. It was fine. And "fine" meant I was settling.
So I kicked off a gathered session. Claude proposed a Busy Parent, a UX Pragmatist, and a Data Architect. I added a Student, a Low-Tech User, and an Empath.
The Busy Parent: "Don't make me explain where I am. You can see where I am. And if I close the chat and come back tomorrow, I need to pick up where I left off. Not start over. I will not start over."
I reacted... of course I reacted. It sparked a thought about how the help-widget-chat pattern is a double-edged sword (people recognize it, but they're trained to ignore it - weird baggage when chat is supposed to be primary). But instead of responding, I noted it for later. The clerk held it. We moved on.
The Student: "This should just be how the app talks to me. Not a feature I find - the surface I live on."
My thoughts swam in two different chat paradigms forming... a "figuring things out" mode and a "feedback" mode. Felt, noted, saved for later. Kept my mouth shut.
The Empath gave voice to what nobody else had: "Is conversation the primary interaction model, or is it supplementary? I don't think you can have it both ways." The UX Pragmatist processed and shared from their perspective: "You absolutely can. You just can't have it both ways at the same time on the same screen."
And somewhere in all of that... in the accumulation, in the not-responding... my thinking shifted. Not from any single voice. From the weight of all of them, absorbed without me arguing back.
By the time I actually spoke, I had to laugh at myself. I'd come in with a whole plan for chat widgets. What came out was completely different. The onboarding isn't a one-time flow, it's a perpetual profile conversation. And the relationship between that conversation and a structured profile dashboard... that's the actual hard problem. Not the one I brought in. Not even close.
Turns out the answer was inside me and my made-up friends the whole time.
The "not yet"
The clerk's synthesis found four areas of clear unity. And one thing that wasn't resolved: the profile page, where structured data and conversation both want to be primary.
Instead of pushing through it, the clerk named it as unresolved and left it for later. Build what's clear. Let the hard problem season.
After all, if me and my made-up friends can't figure out what the profile page should be, why would I spend tokens building another wrong version of it? One good "not yet" prevents a week of building on the wrong paradigm.
Another session (this one)
This blog post ended up going through the same process. Which is... both perfect and ridiculous.
I'd been trying to write this post for weeks. Went through six versions, blind-tested them with reviewers, built a tone guide based on 20 years of my own writing, did quantitative punctuation analysis (I have never once used an em dash in 19 years of writing... but I use the fuck out of ellipsis.). Six rounds of revision and I still wasn't happy with it. The post about deliberation couldn't find its own direction.
So I used the tool on itself. Claude and I convened a committee - a cold reader, a practitioner already in agentic workflows, a writer focused on craft, a reviewer who'd given the previous draft a 1 out of 5 on authenticity. I added a Quaker who was genuinely uncomfortable with the whole premise, and a coder who thought this was a fancy prompt wrapper.
The Quaker: "The renaming is itself an act of extraction... taking a practice that belongs to a community, filing off the name, and shipping it."
The Coder: "You have one anecdote where you came in with a plan and left with a different plan. Was the different plan better?"
I sat with those. Didn't respond. The clerk held them.
Unity came on the structural problem (the post was trying to serve too many audiences and serving none of them well) but not on the deeper tensions. The Quaker's discomfort and the Coder's skepticism didn't resolve. They sat next to each other.
So I walked away. Let it season. Came back days later and the narrative I'd been reaching for was just... there. Not the one I'd been trying to write (explain Quaker process to non-Quakers, explain agentic workflows to non-coders, build a bridge between two worlds nobody asked to be bridged). The actual one. The one about how I make decisions and what I needed AI to help me with.
The post about seasoning got seasoned.[^2]
The point
I want AI to do my dishes. I don't want to hand off the thinking to AI.
Every tool, every workflow, every instinct in this space pushes toward convergence. Pick an option. Ship it. Move on. And most of the time that's right... the velocity is real, the unlock is real, I'm fully bought in. But the decisions that matter most are the ones where I'm the bottleneck, because it's a judgment call or a taste call or a nuanced phiosophical call and I don't have the quick answer for them. For those, speed is the enemy. Not because fast is bad, but because fast skips the part where clarity comes from for me.
This tool doesn't replace me. It makes me do the hard part. Which is what I wanted AI to be for all along.
[^1] I'm barely scratching the surface of over 350 years of practice here. Apologies for the wild oversimplification.
[^2] The irony is not lost on anyone in the room. Including the clerk, who is an AI, and who I'm pretty sure would find this funny if it could.