Meeting Notes · Tuesday, June 2, 2026
With Andrew Gordon Walked Andrew through the strategic pivot from agent-focused to workflow-focused — Flight Plans as atomic, signed, runnable artifacts written in plain English. Andrew sharpened the framing (lead with first principles, not 'contrarian'), pushed for concrete customer examples, and laid out the Dealer Science validation playbook: ten conversations to mature four, with a signed deliverable before the product is final. Black Book came back with a twist — they want us to build their MCP connector. Auto-dealer used-car sourcing remains the leading wedge.
The pivot landed. Andrew read the Flight Plan pitch on screen and the framing held — not “build a better agent,” but “describe a workflow in plain English, ship it as a signed runnable artifact.” The core insight driving the shift: people want work done, not agent interactions. Executive assistants, finance managers, and ops leads don’t want to live in an agent; they want the five-step process to run. The market is oversaturated with “AI” branding — every company from issue trackers to All Birds shoes now claims it — and that saturation is precisely the opening for first-principles automation.
Andrew’s push: concrete examples before broader market approach. Pick 5–10 candidate workflows, solve 3 with a common thread, and let the thread define the wedge. The leading example surfaced from Andrew’s own network — a used-car sourcing manager stitching together auction data, Carfax, Auto iPacket, and market pricing across six disparate sources by hand. Andrew committed to arranging that conversation in the next day or two.
Andrew’s sharper frame on positioning: stop optimizing for contrarian. What matters is first principles and whether anyone will actually use it. The contrarian read emerges from being right about what people need, not from posture. And on validation — straight from his Dealer Science playbook — start ten customer conversations so four mature, with a sales order or letter of intent before the product is finalized. Pitch-deck validation after the build is the failure mode.
Black Book — named as the first connector validation target — came back with a twist: they’re protective of their licensed data and won’t grant direct connector access, but they do want someone to build them an MCP connector / CLI. The contact (progressive thinker in a slower-moving org) is escalating to his boss tomorrow. Competitively, Composio and Executor own the agent-integration space, but both presume an agent-conversation surface. Flight Plans skip that surface entirely.
The strategic pivot
We’re not building agent infrastructure. We’re encapsulating the systems an org already uses, making their capabilities accessible and composable, and shipping the result as a signed artifact written in plain English. The agent disappears. The work happens.
What I put on screen
The pitch Andrew read live during the call:
Describe it. Automate it. Done.
Write a repeatable workflow in plain English, built on your actual systems, runnable on day one. That’s your Flight Plan. Share them and everyone flies faster.
The expanded form:
AI today underserves the people who aren’t engineers, and bogs down the engineers themselves with agent setup, authorization, and integration plumbing that shouldn’t be their problem. A Flight Plan changes both. It encapsulates the systems an org already uses, makes their capabilities accessible and composable, and ships as a signed artifact written in plain English. We believe these primitives compose into a more efficient application of AI — one that reaches more people with real problems.
The signal we keep hearing
“I don’t want to live in an agent.” — feedback from an executive assistant. Danny and Matt have echoed the same sentiment in earlier conversations. The pattern is consistent: people want the work done, not the conversation.
A concrete example — the auto dealer
The walkthrough that anchored the rest of the conversation:
- Pull the deal from the dealership management system
- Open the customer in QuickBooks
- Send the DocuSign loan packet
- Send a welcome SMS
- Schedule the first-payment reminder
Today: a finance manager runs these by hand, or stitches them across half-finished integrations. As an Aileron Capability, they ship as one signed, runnable artifact.
Black Book — the wedge that became a partnership lead
The first connector validation target opened a different door than expected.
Pull the spec, generate the connector, run it through a Flight Plan. Validate the depth-over-breadth thesis on a system used-car desks actually rely on.
Black Book is unwilling to grant direct connector access. Their data is licensed to customers under terms that explicitly preclude new consumption channels they didn’t authorize — a connector we build would create one.
But they’re looking for someone to build them an MCP connector / CLI. The contact reads as a progressive thinker inside a slower-moving organization; he speaks with his boss tomorrow about a follow-up.
Andrew offered to provide air cover from five levels above the contact. ALR declined — at this stage, walking the channels and treating each conversation as exploration keeps the surface honest with the org’s own pace. The Black Book outcome reshapes the connector strategy: the auto-generation thesis still needs a clean validation target, and the partnership-build path is a separate (and potentially earlier-revenue) opportunity.
Competitive landscape
Established presence, broad integration surface. Presumes the agent is the consumer.
Agent-integration focus. Same architectural assumption as Composio.
- Every adjacent solution requires an agent conversation as the surface
- Skills sharing across teams is fragmented and non-atomic
- Technical teams only — not accessible to general users
Flight Plans skip the agent surface. That’s the YC-pitchable position — though Andrew’s refinement: lead with first principles, not “we’re contrarian.” The contrarian read emerges from being right about what people actually need, not from posture.
Technical architecture insights
Flight plans as atomic execution units
- Bundle workflow + connectors + dependencies into one shippable artifact
- Shareable without recipient-side system configuration
- AI handles data extraction and transformation inside the workflow — not as the surface
Connector strategy
GraphQL or OpenAPI in, connector out. The first validation target is the auto-dealer data stack — Black Book named the licensing constraint, so the clean test target needs to be a system whose contract permits new consumption channels.
Depth on systems people actually use beats a wide catalog of shallow integrations.
Some target systems will lack APIs and require web scraping — a real cost line item, not an edge case.
Customer validation — design partners, not the mom test
The methodology debate that anchored the second half of the call.
Andrew’s frame (Dealer Science playbook): start ten customer conversations so four mature. Lead times are long. The mature outcome you’re aiming for is a sales order, letter of intent, or purchase contract — before the product is finalized — with a spelled-out deliverable they agree to pay for. You then build toward that contracted shape.
Project a problem, build a solution, then ask “would you use this?” Get a chorus of enthusiastic yeses with no real purchase intent. Pitch 50 VCs. Get rejected because there’s no validation — and the VCs read it correctly as “you want the check to find your product, not to scale it.”
Engineering-school framing of the same idea — first assignment, day one, is to find a paying design partner. Working relationship is established up front; you solve their problem within the scope of what you believe has broader impact. No end-of-build mismatch where the customer says “actually, no.”
Different methods, identical concept. The trap to avoid: building on projected problems instead of contracted ones. Andrew is not advocating selling vaporware — he’s advocating for getting the agreement-to-pay before the build is closed out, so the build aims at a real shape someone has already committed to.
The vaporware constraint
ALR’s hard line: you can’t put nothing in someone’s hands. The “negative-one-to-zero” stretch — getting a runnable thing back up so each customer call can move from “describe the workflow” to “watch a version of it run” — is the gating step. The system was working better two weeks ago than it is today; the rebuild is now blocking customer conversations, not just internal velocity.
Go-to-market
Andrew’s framing: pick 5–10 concrete candidate workflows, solve 3 with a common thread, and let the thread define the wedge. Don’t chase a broad market message until concrete examples earn the right to generalize.
The positioning shift that follows from the pivot:
- Target non-technical users, not the current technical audience
- Positioning anchor: “We’re not building better agents — we’re building better automation.” Lead with first principles, not with the contrarian framing itself
- Y Combinator pitch angle identified; needs concrete customer examples and at least one contracted design partner before pitching
Use cases surfaced
Needs: auction data + Carfax + Auto iPacket + market pricing, fused into a sourcing decision.
Today: manual gathering across six disparate sources.
Challenge: some sources have no API — scraping required.
Andrew Gordon’s direct contact; meeting to be arranged today or tomorrow.
Audit scheduled hours in ADP against actual hours worked. Simpler integration profile and a clear compliance use case.
Next steps
It was working better two weeks ago than it is now. Get it back operational so every customer conversation can move from “describe the workflow” to “watch a version run.” This is now the gate on the customer-conversation cadence.
Used-car sourcing manager. Today or tomorrow. Goal: understand the specific workflow requirements before we build to them.
Contact escalates internally tomorrow. Re-engage with two possible shapes on the table: a build-for-Black-Book MCP connector / CLI engagement, and clarity on whether a connector for consuming their data is fully off the table.
Black Book’s licensing rules out the connector auto-generation test there. Pick a system whose contract permits new consumption channels and run the same experiment — depth-over-breadth needs at least one clean validation.
Enumerate candidate Flight Plans across auto-dealer, HR, and adjacent domains. Find the common thread that defines the wedge.
One sales order or letter of intent with a spelled-out deliverable. That’s the validation Andrew is pushing toward and the gate on any later fundraise conversation.