Parrot (YC Spring 2026): teardown vs XMarts

In one line: Parrot sells an AI "back end" to US collision repair shops: agents that audit estimates, chase insurers, call parts suppliers and answer the phone, sitting on top of the shop's existing CCC ONE software [site] [yc]. Threat: 2, because it is the closest live example of our W5 shape (one layer for one type of small firm), but it has two people, two claimed customers and a different industry [yc] [press].

1. Fact file

Item	Finding	Source
What they say	Site: "AI coworker for collision repair shops". YC line: "Agentic operating system for auto repair shops".	[site] [yc]
Founded, team, funding	Founded 2026. Two people. San Francisco on the YC page, Austin, Texas on the site footer (they disagree). Funding: UNVERIFIED. One aggregator lists $125k, another $500k, neither has a primary source.	[yc] [site] [press]
Founders	Phillip Baek (CEO): wrote lab-automation software as a Stanford high-school researcher, interned at Amazon on workflow automation for old-style businesses. Liam McBride (CTO): dropped out of a UIUC master's in computer engineering, applied machine learning. Their last startup automated image-analysis workflows for biomedical researchers. What YC likely saw: two builders who automate messy workflows and moved fast to a new market (inference).	[yc] [press]
Customers	They claim two independent auto shops signed at $30K a year each (launch post, three weeks after the pivot). The site names one owner quote (Apex Auto Collision, Austin) and shows ten Google reviews of Austin customers, which are reviews of the shop's repairs, not of Parrot (inference from the page layout as described).	[yc] [site]
Price	Public price: not found (/pricing returned 404 on 2026-10-19). Signed contracts at $30K a year each (they claim). No job ads (YC page shows 0 open roles; /careers returned 404).	[yc] [jobs]
Path	Alchemy (bioimaging workflow app, same two founders, also listed as YC Spring 2026) → Parrot, "auto repair OS" (launch post) → narrowed to collision repair on the site. They say they pivoted three weeks before the launch post. Press report dated 11 Oct 2026 repeats this and cites a launch report of 2 June.	[yc] [press]

2. Their proposition, reverse engineered

Part	What they do
Who buys	The independent collision shop owner or manager. Test customers are in Austin, Texas [site] [yc].
The pain	Shops run on calls, texts and emails to insurers, parts suppliers and customers. YC launch post: the work "depends on phone calls, messages, and emails" and only recently became automatable [yc].
The promise	"Estimates, parts, claims, and taxes, handled end to end" [site]. They claim "$15K additional monthly revenue", "23% more customers", "5x manager efficiency" [site]. No method or named shop is given.
How it works	Download a Mac or Windows app, connect CCC ONE, phone, text, email and calendar, then it audits estimates, chases claims and answers calls [site]. Works with CCC ONE, Audatex and Mitchell 1 [site]. Early access: setup is by booked call, not self-serve [site].
The output	An estimate audit that finds missed operations and hidden damage. Supplier quote comparisons from voice-agent calls. Scheduled insurer follow-ups until a claim is approved. Invoice matching, tax tracking, booked appointments [site].
Proof style	Revenue and customer-count percentages, one owner quote, signed-contract count and size [site] [yc]. No hours or jobs counted in public.
Who sells it	Founders, by demo call, working in shops themselves [site] [yc].
Price model	Not public. Two signed deals at about $30K a year (they claim) [yc].
The story	Enemy: shops still run on calls and paperwork. Belief: auto repair is the next trade to go from record-keeping tools to an operating system, like QuickBooks to ServiceTitan or Micros to Toast [yc].

3. Positioning

Category they claim: "Agentic operating system" / "AI coworker" for repair shops [yc] [site].
Category a YC partner would put them in: vertical AI for one trade, with voice agents as the wedge (inference). Same family as Rulebase-style full-loop vertical agents (brief).
What they want to be known for: the AI that does the phone work between a shop and its insurers and suppliers.
Where they sit on the map: L1/L2/L3 = 2/2/1. Buyer = sme, business.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (Marton's logs show)
L1 Company brain	2. They say the product "maintains context" across estimates, parts, customers, insurers, suppliers and payments [yc]. It reads the shop's own system (CCC ONE) and calendar [site]. Nothing public on how it learns over time.	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning. Built for one person, on Macs.
L2 Agent orchestration	2. Several jobs by agents (calls to suppliers, adjuster chasing, inbound phone, job assignment) [site]. Approval and handover rules are not described: UNVERIFIED.	About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests. Only about 15% of jobs reached the main code, and merges were unreliable in October.
L3 Execution and learning	1. It does the work (calls, chasing, audits). Its checks are on other people's work (estimate audit, invoice match) [site]. No public way the results change the next run. Claimed revenue numbers have no method [site].	About 160 failure records, each with an automatic check. Offer feedback lines fell from 44 to 18 as 22 rules were added. No outside customer yet.

5. Same and different

Topic	Same or different	Them	Us
Core shape	Same	One agent layer for one type of small firm, with voice agents that call real people [yc].	Wedge W5 is the same shape (brief).
Belief	Same	Work between systems and people can now be done by agents, not just logged [yc] [press].	Same belief (brief).
Kind of work	Different	Phone and paperwork for shops: insurers, parts, customers [site].	Not set. Our wedges are code, memory, margin and skills (brief).
Buyer	Different	US collision shop owners [site] [yc].	SMEs and scale-ups, UK first (brief). No customer yet.
Delivery	Different	Desktop app on the shop's own computer, inside CCC ONE [site].	A system the founders run with the client (brief).
Owner of the result	Same	The agents do the task, a person stays the owner (inference).	Same (inference from the brief).
Proof	Different	Percentages with no method, two signed deals [site] [yc].	No customer number yet. Own-business numbers only.
Speed to start	Different	Setup by booked call, early access [site].	Not set.
Price	Different	About $30K a year per shop (they claim) [yc].	Not set.
Size	Same	Two founders [yc].	Two founders (brief).

6. Wedge overlap

Wedge	Fit	Why
W5 One layer for one type of small firm	direct	Same shape: one trade, an agent layer on its existing software, sold to the owner [yc] [site]. The trade differs (collision repair, not accountancy), so this is a pattern rival, not a sector rival (inference).
W7 Find and fix margin leaks	adjacent	Estimate audits find missed operations and invoice matching catches mismatches, which recovers money for the owner [site]. They do not say they measure margin or find pricing versus demand.

7. Whitespace (where they stop)

Layer gap: no public sign that results change how the next run works [site] (inference from silence). To own it we need a rule-change log tied to checked runs, with rework falling.
Buyer gap: US collision shops only; integrations are CCC ONE, Audatex, Mitchell 1 [site]. Nothing for UK firms or other trades. To own it we need a named UK buyer type with the pain on file.
Number gap: claimed revenue and customer gains, no method, no named shop [site]. To own it we need one before/after number a finance person would sign, with the method shown.
Price gap: about $30K a year per shop (they claim) [yc]. A very small firm may not pay that (inference). To own it we need a price a 10 to 50 person firm can pay, tied to the result.
Speed gap: not a gap we can claim. They are quick to sign, and setup is a booked call [site] [yc]. Skip.

8. For the YC application

Proven today: Marton's logs show an agent system that runs 24/7 and checks its own work with a second AI, with about 160 failure records that each added an automatic check. Parrot's public proof is two claimed contracts and unsourced percentages [yc] [site].
Draft answer to "how are you different from Parrot?": DRAFT. "Parrot built an agent layer for one trade, US collision shops. We run the full loop on our own business today: agents do the work, a second AI checks it, and every failure adds a rule." Founders say it their own way.
Future claim (needs proof): "We own one number for one type of UK firm, and the system gets better from checked results." Needs one customer workflow run for 4+ weeks with a before/after number and a rule-change log.
What we understand that they don't: a layer for one trade is only worth buying if the result is checked and measured, so the agents improve from it. needs proof on a customer; Marton's logs show the loop on his own work (offer feedback lines fell from 44 to 18).
Copyability (M3): medium. A learning loop from checked results is months of work and needs shop data over time (inference). They could add it, and it fits their buyer. But two people are busy selling and building integrations in one US trade, and a UK or multi-trade move would pull them off it (inference from [yc] and [site]). Not hard, so M3 can not score above 3 against Parrot alone.
Why now (M4): None found in this teardown. Their own launch post says calls and messages only recently became automatable, with no date or source [yc]. Do not use it.
Do not claim: "no one is doing a vertical agent layer for a small trade" (Parrot, and CarSignal in the same trade, say otherwise). Any outcome number for a customer. "Learns from checked results" as a claim about a customer.

9. Ideas worth taking

Sit inside the software the buyer already pays for (CCC ONE) instead of replacing it [site]. It cuts the switching cost.
Founders work in the customer's shop and sign before they build [yc]. Fast proof of willingness to pay.
A single launch line that tells the story of the trade: QuickBooks to ServiceTitan, Micros to Toast [yc]. A clear way to make a boring market feel inevitable.

10. Unknowns

Real price, plans and contract terms. /pricing returned 404. Ask a shop that has seen a quote.
Whether the two $30K deals are paid, live or pilots. Only the founders' own post says it.
Who the second customer is, and whether the "20% revenue" and "5 days to 2 days" claims on a third-party page are theirs. That page names no shop or method [press]. Treat as UNVERIFIED.
Funding raised. Two aggregators disagree.
Where they are based (San Francisco on YC, Austin on the site).
Whether they sell outside the US. Check careers (404 now) and any UK page later.
How the agents are approved or overruled by the shop, and how the product learns. Check the help center (404 now) when it goes live.
Related rival in the same trade: CarSignal (YC, "The AI Operating System for Auto Shops"), which claims 40+ shops live across 18 states (they claim) [yc-carsignal]. It is a different angle (diagnostics and the full repair job). It should get its own line in Pass 3.

Sources

useparrot.com [site]
YC company page: Parrot [yc]
Launch YC: Parrot, AI-native OS for auto repair shops [yc]
Startup Spotlight: Parrot pivoted into collision repair and signed two $30,000 shop contracts, 11 Oct 2026 [press]
Alchemy (same founders), Zeitgeist profile [press]
Extruct profile of useparrot.com [press], unverified claims
The YC Tier List: Parrot [press], third-party opinion only
YC company page: CarSignal [yc-carsignal]
Jobs: [jobs] YC page shows 0 open roles; /careers, /pricing, /changelog, /help and /security on useparrot.com returned 404 on 2026-10-19.

Parrot

What they do

How close to us

Wedges

Who buys

Full teardown