Meeting Notes · Friday, June 5, 2026
With Andrew Gordon ALR walked Andrew through the narrowed thesis — workflows authored, shared, and forked across an org instead of chatbot-per-person — and Andrew responded by laying out his own auto-dealer 'meta-skill' concept (MCP → API → browser → desktop fallback hierarchy, systems of action layered on systems of record). The two pitches landed on the same substrate from opposite directions. Andrew opened three intros — Mari at Intuit, Justin Lee at Automotive Ventures, Tomasz Tunguz at Theory Ventures — and brought a sharp productization frame from one of his mentors: stop selling the grocery list, sell the pizza. He also left the door open on operational involvement.
A check-in call that became a two-way feedback session. ALR opened with the narrowed thesis — the super-user productivity gap is an organizational rollout problem, and chatbot-per-person is the wrong surface; authored workflows that can be shared, forked, and compounded across an org are the right one. Andrew then offered his own work-in-progress for pressure-testing: a meta-skill for auto dealers that builds the integration it needs (MCP if it exists, API if not, browser-use as fallback, desktop infra if forced), with the goal of stacking systems of action on top of the dealership’s bespoke systems of record. Mid-call ALR pointed out that Andrew had arrived at the same substrate ALR opened with — connectors as primitives, skills layered on top, integrations that compound — just from the auto-dealer direction.
The most actionable output: three intros. Mari at Intuit — a manager running AI upskilling at Intuit who, in Andrew’s words, was “literally complaining about the problem you’re solving the other day.” On vacation but open to a text. Justin Lee at Automotive Ventures — VC intro Andrew offered to make on the spot. Tomasz Tunguz at Theory Ventures — VC publishing about a frontier-model-creates-skills / local-model-executes-skills pattern that mirrors the Aileron architecture. ALR also surfaced Brandon at Kapor Capital as a parallel signal — Brandon’s partner is hand-rolling skills plus an AI-built web dashboard to share them. VCs may be an attractive beachhead: minimal system-integration complexity, outsized influence on downstream adoption.
The framing that stuck was from one of Andrew’s mentors: “Some people just want to order pizza.” The thesis is right, but selling ingredients (or even a grocery list, or even a meal kit) isn’t the unlock — packaging the value as something pre-prepared is. For auto dealers specifically: don’t sell “the dealership of the future” — sell “prep this service appointment and book a 5–10% revenue bump.”
Andrew also left the door open on operational involvement. He’s not stepping in today — he’s heads-down on the auto-dealer concept himself — but the framing was “maybe we work together, maybe one of us does something and the other doesn’t, maybe one comes to throw in on the other.” He’d invest if given the opportunity. The path stays warm.
Where ALR is at
- Talk to as many people as possible — warm network first, then targeted outreach designed to yield quality conversations rather than volume
- Write an investment memo and supporting blog posts, now that the thesis is narrow enough to defend with research
- Accept the discomfort of the early-stage network-mining phase — customer discovery is bound by who you can reach, and there’s no way around it
The narrowed thesis
ALR re-pitched the through-line, sharpened from the prior weeks of conversation.
AI-native individuals are getting roughly 5x productivity gains. Organizations aren’t seeing meaningful ROI because the people getting those gains are individuals, and there’s no path to roll the gains out. The unstated ask we’re putting on every employee — learn AI, learn to co-work with Claude, configure your own setup — is the bottleneck.
Will Dellwo said exactly the same thing in his own discovery call. The same pattern is showing up across every conversation.
What goes in place of chatbot-per-person
Open-ended, infinitely flexible, and therefore puts the entire burden of configuration and use onto the operator. Personal, not organizational. Skills (in the Claude sense) are locally installed and not portable.
- A workflow you author once, fire many times
- Structured inputs along the way — and a UI smart enough to render itself for each phase
- Shareable across the org, forkable by the next user, improved upon and re-shared
- Connected to your actual systems through services, not just notes in a chat history
- Compounding gains across the org, not stranded in one person’s window
The Lassie shape
ALR brought back the Lassie reference from last week — start from an industry you know (in their case, dental), build for that ICP, but think generally from the start so the wedge expands. “The beachhead market still exists. The wedge still exists. But you have to be thinking generally from the start.”
Halfway through the call Andrew flipped the agenda and walked through the concept he’s actively building toward, asking for pressure-testing.
The auto-dealer ecosystem is built on bespoke tools that don’t share data or integrate. There are huge data-ownership and access issues — different rules apply if you’re a dealer vs. a third party. The result: no innovation surface, because the prerequisite plumbing isn’t there.
The integration fallback hierarchy
The “meta-skill” is a skill that builds an integration on demand, walking down a fallback ladder until something works:
If the system exposes an MCP server, use it. If not, fall back to whatever first-party API exists.
No API? Drive the browser. Still won’t work? Browser-use. As a last resort, a desktop with whatever runtime infrastructure is needed to make the legacy interface yield.
The example: service-appointment prep
- CRM — Andy has a 9am tomorrow, third year servicing here
- Service history — last visit flagged tires getting low, not ready then but close
- Recall data — open recall on this VIN
- Tire monitoring / inventory — current condition, parts available
Today: good service departments do this fusion by hand, and most don’t do it at all. Done agentically — and surfaced either to the service advisor for prep, or proactively to the customer via the dealer’s CRM (“we noticed an open recall, want us to add it to your appointment?”) — Andrew’s conservative estimate is a 5–10% revenue bump at the service department.
Systems of action, layered on systems of record
The deeper architectural bet: once these integrations exist, the dealer’s bespoke tools become systems of record, and a thin layer above them — Aileron-shaped — becomes the system of action. The old UIs and clunky vendor portals stop being the interface for getting work done; they become storage. Whether the customer engages via text, agent, or human is a presentation question, not an architecture question.
The integration math: today every new use case is m × n because each tool needs to talk to each other tool. Build the connectors and skills as tool calls and it collapses to m + n — each new integration compounds across every skill that already exists.
The convergence
Mid-call, ALR named what was happening. Andrew opened with “agents in dealerships” and ended up describing connectors as primitives, skills layered on top, integrations that compound, and a substrate that detaches the work from the legacy UIs. That’s the same architecture ALR has been pitching for weeks — from the auto-dealer direction instead of the org-rollout direction.
“It feels like we’ve just flip-flopped. You’re now talking to me in the terms where it seems like you’re recognizing that a lot of the value is derived from connecting into these other external services. Those things kind of become primitives. And then when you start to build them together, you’ve unlocked a whole bunch of potential that you don’t necessarily need to plan for or fully understand. What you’ve done is created the substrate.” — ALR
Andrew agreed. The convergence is a useful signal — two people working on adjacent surfaces independently land on the same primitives.
‘Pizza, not ingredients’
The productization frame Andrew brought, from one of his mentors:
“Some people just want to order pizza.”
The thesis can be right and the customer can still bounce off it because you’re handing them ingredients (or even a grocery list, or even a meal kit) when what they need is the pizza. For Aileron and for Andrew’s auto-dealer concept: don’t pitch the substrate, the architecture, or “the dealership of the future.” Pitch the specific revenue-bearing outcome that the substrate makes possible.
The corollary on customer pain: dealers don’t feel the pain until they see the solution. “They’re not going to see the pain. But they will say ‘I don’t know where to start on AI.’” The job is to convert that vague unease into a concrete demo that makes the cost of doing it the old way suddenly visible.
Intros and signals
Manager at Intuit leading a team that’s actively figuring out how to upskill into AI and co-work with it. In Andrew’s words, “literally complaining about the problem you’re solving the other day.” On vacation this week but open to a text — Andrew suggested ALR text her directly: “hey, heard you’re on vacation, want to grab a call when you’re back / during downtime / whatever works.” Texted intro already in flight.
Andrew’s character read: “honest and straightforward without being mean.”
VC focused on automotive. Andrew offered the intro on the spot. Pairs naturally with the auto-dealer thread that’s already running through Black Book, Matt Blatt Kia, and the used-car sourcing manager.
VC publishing about the same architectural pattern from inside the VC workflow: frontier model creates the skills, local models execute them, applied to company research and assembly. Andrew’s read: “VCs pass around his shit not knowing how to do it all the time.”
Link: tomtunguz.com
Impact-investing VC. ALR met with him yesterday — Brandon wants to use AI to be more efficient with his research, and his partner is hand-rolling skills plus an AI-built web dashboard to share them across the firm. Same shape as the Tunguz signal: VCs reaching for skill-sharing infrastructure that doesn’t yet exist as a product.
Why VCs may be the right beachhead
VCs aren’t running into massive bespoke enterprise systems. The integration surface is small: research, knowledge management, light CRM. But the influence is outsized — every portfolio company watches what their VCs are using. If the partners at a respected firm are forking each other’s skills in an Aileron-shaped substrate, that signal travels.
Worth threading the VC angle through Justin Lee, Tomasz Tunguz, and Brandon as a coordinated pass rather than three separate conversations.
Andrew’s relationship to the work
ALR asked directly whether Andrew wanted to get operationally involved.
Not stepping in operationally right now — he’s heads-down on the auto-dealer concept and that’s his current sandbox. But the framing was open: “Maybe we work together. Maybe one of us does something and the other doesn’t. Maybe one comes to throw in on the other. I don’t know.” He also said, unprompted, that he’d invest if given the opportunity and would pressure-test the business plan as a friend.
The two threads — ALR’s substrate and Andrew’s auto-dealer meta-skill — are close enough architecturally that the door stays open in both directions. Don’t force the frame; keep talking.
Next steps
Use Andrew’s suggested low-pressure framing — “heard you’re on vacation, happy to find time when works for you.” Don’t pitch on the first message.
Pair with the existing auto-dealer thread (Black Book, Matt Blatt Kia, used-car sourcing).
Specifically the frontier-creates / local-executes pattern. Cross-reference against Aileron’s actual architecture.
A self-built version of the thing Aileron is building — strong signal about both the need and the current pain of doing it by hand.
Andrew had to drop. Continuation planned for later in the day once his kids are down — separate thread on pressure-testing the auto-dealer technical thesis specifically.
The thesis is narrow enough to defend. Get the written version out so the next round of conversations starts from a sharper anchor than the current intro can carry.