MSPilot (YC Fall 2025): teardown vs XMarts

In one line: MSPilot sells IT service firms (MSPs) a governed AI workspace that they resell to their small-business clients at a monthly seat price, with the MSP keeping the margin [site]. Threat: 1, because it shares our small-firm buyer world but sells a governance and billing layer, not finished work [site]. It matters more as a channel and a pricing signal than as a rival.

1. Fact file

Item	Finding	Source
What they say	"Turn Shadow AI into your next revenue line." YC line: "Turn AI adoption into a recurring revenue line." About page: "The AI control plane for the MSP."	[site] [yc]
Founded, team, funding	YC Fall 2025. The YC page says team of 3 and "Founded 2026" (as listed). The About page names 8 people. Funding: UNVERIFIED. A directory lists a $125k seed but shows no investor or date. The About page says YC and Focal Ventures back it; I could not find Focal's side of that.	[yc] [site] [press]
Founders	Paul Li (3x founder, Waterloo CS, ships coding tools they claim 1.5M developers installed) and James Jiang (ex-Amazon developer). They met at Waterloo and have known each other 10+ years. YC likely saw a repeat builder pair with a real distribution record (inference).	[yc]
Customers	11 logos on the home page, including Brightworks, Datatel, Integrita Systems, Layer 2 Computers, Solved IT Canada, Australian Cyber Defence. The About page calls them "early design partners", led by Zinck Computer Group. No case study, no revenue number. A job ad claims "paying customers" but names none.	[site] [jobs]
Price	Public. Standard seat $10 per user per month, locked to one model. Power seat: pay for tokens at cost, no minimum. Month to month, no lock-in. The MSP resells at an expected $20 to $30 per user.	[site]
Path	Four names: Personabo, EasyCode (a coding assistant), OpenBuilder (open-source "Lovable" alternative, YC launch post), then MSPilot. The launch post showed "11 months ago" when read in Oct 2026, so about Nov 2025 (inference). Same two founders throughout.	[yc]

2. Their proposition, reverse engineered

Part	What they do
Who buys	The MSP owner. The end users are the MSP's small-business clients [site].
The pain	Staff at client firms already use ChatGPT, Claude and Gemini on their own. The MSP has no control and no revenue from it ("shadow AI") [site].
The promise	Resell governed AI apps, keep the data safe, bill monthly, and keep the margin [site].
How it works	Pushes a governed desktop workspace to every device through the MSP's own RMM tools. Seats and usage flow through the PSA into the invoices the MSP already sends. Connects to 8 apps (Drive, SharePoint, Slack, Salesforce, Notion, HubSpot, QuickBooks, Dropbox) through governed MCP connectors. Blocks or redacts sensitive data before it reaches a model [site].
The output	A branded AI workspace per client, an audit log, and a monthly bill line. Next step they describe: usage analytics flag repeated prompts as candidates for a billable custom agent [site].
Proof style	Counts of features and logos. A profit calculator: 25 clients, 250 staff, $22 seat, 45% margin, $30,000 a year (illustrative, they say results vary) [site]. They also claim client demand for AI agents rose from 1% to 7% in six months [press].
Who sells it	Founders and a founding account executive, no SDR hand-off. A partner portal behind a password for early-access partners [site] [jobs].
Price model	Per seat per month plus token pass-through [site].
The story	AI is a recurring revenue line for the MSP, not a threat. Treat AI as a business conversation, not only an IT one [site].

3. Positioning

Category they claim: "AI control plane for the MSP" [site]. Job ad: "blue-ocean" category with no incumbent [jobs].
Category a YC partner would put them in: a reseller platform for AI seats, with governance as the hook (inference). Close to what a security vendor for MSPs sells [press].
What they want to be known for: the easiest way for an MSP to sell and bill AI.
Where they sit on the map: layers L1/L2/L3 = 1/1/1, buyer = sme, business.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (Marton's logs show)
L1 Company brain	1. A shared knowledge base of skills, connectors and instructions, scoped to all clients, one client or one team. Imports chat history. No learning or lasting memory described [site].	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning.
L2 Agent orchestration	1. Shipped today: governed chat apps and MCP connectors with permissions. A runtime with approvals, vaulted credentials and a template builder appears only as work to be built in a July 2026 job ad [site] [jobs].	About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests.
L3 Execution and learning	1. Meters, logs and reports AI use per client. It does not do the client's work or measure a result: "checks only" [site].	About 160 failure records, each with an automatic check. Offer feedback lines fell from 44 to 18 as 22 rules were added. Only about 15% of jobs reached the main code, and merges were unreliable in October.

5. Same and different

Topic	Same or different	Them	Us
Core belief	Different	Staff use AI anyway, so govern it and bill for it [site].	Agents should finish the work and the brain should learn from checked results (brief).
Buyer	Different	The MSP owner, who resells to small firms [site].	SMEs and scale-ups directly (brief). No customer yet.
Owner of the result	Different	The MSP and its client own whatever gets built [site].	We would own one number on one workflow (needs proof).
Kind of work	Different	Chat access, data control, billing [site].	Marton's logs show agents do and check the work, on his own business only.
Skills and shared knowledge	Same	A shared base of skills and instructions rolled out to all users [site].	Marton's logs show 98 skills and a vault served to agents.
Tooling around agents	Same	Audit trail on every connector call, drift detection on devices [site] [jobs].	Marton's logs show a heartbeat, failure records and cost tracking.
Delivery	Different	Software through existing RMM and PSA tools [site].	The system plus the two founders.
Proof	Different	Logos and a calculator, no customer result [site].	No customer number yet.
Price	Different	$10 seat plus tokens, public [site].	Not set yet.

6. Wedge overlap

Wedge	Fit	Why
W2 Seeing whether your AI agents actually produce	adjacent	They log AI use, cost and policy hits per client [site]. They show who used AI, not whether work got done. Buyer is an MSP, not a team running agents.
W5 One layer for one type of small firm	adjacent	Same shape: one layer built once for one firm type, sold many times. The firm type is the MSP and the layer is governance and billing, not an operating layer for the firm's own work [site]. The screen marked this direct; I rate it adjacent because the problem and the buyer differ.
W8 The other 90% of an AI skill	adjacent	They roll out shared skills and instructions from one console [site]. No public claim of upkeep from real use (UNVERIFIED that none exists).

7. Whitespace (where they stop)

Layer gap: they control and meter AI use; they do not do the work or check it [site]. To own it we need one client workflow run by our agents for 4+ weeks.
Buyer gap: they sell to MSPs. Small firms with no MSP, or that want a result and not a seat, are outside the product (inference). To own it we need a small firm saying "I want this done, not a tool".
Number gap: the only number is an illustrative margin on the MSP's side, with no client result [site]. To own it we need a before/after number a finance person would sign.
Price gap: not a gap. $10 a seat is cheap and public [site].
Speed gap: not shown. They deploy through the MSP's RMM tools [site], but no time to first value is given.

8. For the YC application

Proven today: Marton's logs show his system already runs agents that do and check work: about 5,600 jobs, each checked by a different AI, with about 160 failure records that each added an automatic check. MSPilot's public product governs and bills AI use; it does not claim to finish a client's work [site].
Draft answer to "how are you different from MSPilot?": DRAFT. "MSPilot helps IT firms sell AI seats and keep the data safe. We run agents that do one job end to end and check it, and we already do that every day on our own business." Founders say it their own way.
Future claim (needs proof): "A small firm pays for a result, not a seat." Needs one client workflow run by our agents for 4+ weeks with a before/after number.
What we understand that they don't: governance tells you who used AI; checked results tell you whether work got done. needs proof on a customer; Marton's logs show it on his own work.
Copyability (M3): medium. They already plan an agent runtime with approvals and a template builder [jobs], and they flag repeated prompts as agent candidates [site]. Building a loop that learns from checked results would pull them away from selling seats through RMM tools (inference). Not hard, so M3 can not score above 3 against MSPilot alone.
Why now (M4): None found in this teardown. Do not use general AI progress.
Do not claim: that no one governs agents (DefensX and dope.security sell governance to MSPs [press]); that small firms want agents from us (no calls yet); that MSPilot has no traction (UNVERIFIED either way).

9. Ideas worth taking

Give the buyer one number on the first page. Their margin calculator is a clear, simple hook. We need a real before/after number to do the same.
Reach the buyer through tools already in their stack (RMM, PSA). It cuts setup time.
Find the next agent from real use: usage analytics flag repeated prompts and documents. This is the same idea as W8 skill upkeep, started from the buyer's data.

10. Unknowns

Whether the 11 logos pay. Look for a price or invoice mention, or ask an MSP who has seen a demo.
Funding, revenue and customer count. Check the YC directory notes, Crunchbase and the founders' public posts. (Not LinkedIn.)
Whether the agent runtime has shipped. The only evidence is a July 2026 job ad marked inactive [jobs].
Whether Focal Ventures invested. The About page says so [site]; two searches found no confirmation.
The home page calculator shows a $500 MSPilot fee for 250 staff ($2 a seat) while the pricing page shows $10 a seat. They do not match on their face; the pricing page may include model cost (UNVERIFIED reading).
UK or European reach. Customers shown are US, Canada and Australia; pricing is in dollars. Unknown.

Sources

mspilot.io [site]
mspilot.io/pricing [site]
mspilot.io/about [site]
mspilot.io/features [site]
mspilot.io/how-to-sell-ai [site]
mspilot.io/partners [site]
YC company page [yc]
YC launch post: OpenBuilder [yc]
Founding Engineer, AI Infra job ad on DOU, 6 Jul 2026 [jobs]
YC jobs page for MSPilot, as listed on the YC company page (Founding Engineer, Founding GTM Specialist) [jobs]
Dealroom profile [press]
Extruct profile [press]
DefensX: MSPs new revenue stream, secure AI workspace [press]
dope.security: shadow AI tools for MSPs [press]

MSPilot

What they do

How close to us

Wedges

Who buys

Full teardown