Weave (YC Winter 2025): teardown vs XMarts

In one line: Weave reads a company's pull requests to measure how much engineering work gets done, how much AI wrote, and what AI costs, then offers a router that sends coding-agent requests to cheaper models [site] [yc]. Threat: 3, because it sells to the same buyer on the same problem as W2 (are my AI agents producing, and at what cost), it is funded and live, and it names customers [site] [press].

1. Fact file

Item	Finding	Source
What they say	Home page: "Turn your software team into a software factory". YC line: "AI to understand and then route engineering work".	[site] [yc]
Founded, team, funding	Founded 2024, 16 people, San Francisco [yc]. They report a $13.5M Series A on 28 Jul 2026, led by Standard Capital, with YC, Moonfire, Burst Capital and others [press]. A $4.2M seed in Jul 2025 is reported in search results only (UNVERIFIED on a primary page) [press]. Total about $17.7M (inference: 13.5 + 4.2).	[yc] [press]
Founders	Adam Cohen (CEO): operations and sales, ran an organisation of 100+ people, ran sales and customer success at Causal. Andrew Churchill (CTO): employee #1 at Causal, MIT computer science and maths. They met at Causal. What YC likely saw: a seller who wants engineering judged on a hard number, plus a technical co-founder who built a product at an earlier startup (inference).	[yc] [press]
Customers	They claim 500+ engineering organisations. Named: PostHog, Reducto, Bytes AI, Telnyx, Intellistack, Remi, OpenZeppelin, Robinhood, Sonatype, Extra Space Storage, WELL Health, JAGGAER, Redwood. Counts differ by page (20K on the customers page, 25K+ on the home page; 22.6K not re-found), so treat all as claims.	[site] [yc] [blog]
Price	Public. Starter $0 (the "aimed at teams under 5" part is UNVERIFIED; the page says only "free to get started"). Pro $50 per engineer per month. Enterprise custom. Router: 5% of routed cost for solo developers and startups; teams of 50+ are sent to a demo.	[site] [blog]
Path	Sold as "WorkWeave" (the old web address workweave.dev still appears). Now "Weave" on weaveos.com. No dated rename post found. Core product: measure PR output ("Weave Hours"). Added AI-impact reports, then Agent Observability, then the Router (2.0 on 9 Sep 2026).	[site] [blog] [press]

2. Their proposition, reverse engineered

Part	What they do
Who buys	Engineering leaders. Startups under 5 engineers up to enterprises such as Robinhood [site]. Enterprise page targets "complex engineering orgs" [site].
The pain	Engineering is judged "on vibes" while sales has one hard number (CEO, press) [press]. Leaders cannot tell if AI spend pays off [press] [blog].
The promise	Show how much faster AI makes the team, the effect on quality and the cost. Then cut model cost without losing quality [yc] [site].
How it works	Connect GitHub. Models score every PR and review for size and quality [press]. Router: one install command, then requests are routed by difficulty [site] [blog].
The output	Dashboards and reports: output score, AI ROI, DORA, token spend per output, benchmarks vs other orgs. A chat agent called Wooly. A router endpoint. Q3 AI Impact Report [site] [blog].
Proof style	Self-reported numbers: Reducto +19% output in a measured period, Bytes AI about 30% in six weeks, OpenZeppelin saved four weeks, 14 days from install to first executive report [site]. Router tests on public benchmarks, with overlapping confidence intervals admitted [blog].
Who sells it	Free self-serve tier plus demo. Open roles: founding account executive ($200K to $350K) and forward deployed engineer [jobs]. Open-source router repo (5.6k stars, they claim) [site].
Price model	Per engineer per month for analytics; a percentage of routed spend for the router [site] [blog].
The story	Their enemy is judging engineering by gut feel. Their belief: engineering output can be measured like revenue, and that is the system of record for people and AI agents together [press] [site].

3. Positioning

Category they claim: engineering intelligence / AI impact and ROI, plus a router [site].
Category a YC partner would put them in: developer productivity analytics (the DX, Jellyfish, LinearB set; their own Compare page lists those names) [site], with an AI cost-control add-on (inference).
What they want to be known for: the system of record for how engineering teams and AI agents affect business results [press].
Where they sit on the map: layers L1/L2/L3 = 1/1/1, buyer = mid-market, business (engineering leader).

4. Three layers vs ours

Layer	Them (score + evidence)	Us (from the system inventory, "Marton's logs show")
L1 Company brain	1. Builds a model of engineering work from code data and team feedback. It is not a memory of the whole company that other work reads. [site] [blog]	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning.
L2 Agent orchestration	1. The Router sits in front of coding agents and picks models; an escalation classifier moves stalled tasks to a stronger model [blog]. The Wooly agent answers questions about the data [site]. It does not run a team of agents with approvals or handovers.	About 5,600 jobs run by agents from three makers, each built in its own copy, checked by a different AI, then automatic tests.
L3 Execution and learning	1. Measures and routes; it does not do the work (checks only). The router learns from team feedback and session state [site] [blog], which is a real loop on the routing choice, so a 2 is arguable (inference). The screen pass gave 2; the deep pass keeps 1 because the work itself is done by someone else's agent.	Every failure gets a written record and an automatic check (about 160 records). Offer feedback lines fell from 44 to 18 as 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	Same	AI agents and engineers need to be measured on output, quality and cost [site].	Marton's logs show the system tracks what is stuck, what silently stopped and what it costs.
Buyer	Same	Head of engineering, from small startups upward [site].	W1 and W2 target small engineering teams and teams already running coding agents (brief). No customer yet.
Vantage point	Different	Outside view: reads PRs, reviews, spend after the fact [press].	Inside view: the same system queues, builds, reviews and lands the work (Marton's logs show).
Who does the work	Different	Someone else's agents and people. Weave measures and routes [site].	Our agents do and check the work, on Marton's own business only.
Kind of number	Different	Modelled "expert hours" from PR analysis. Their own Q3 post says it is not recorded hours saved and not proof AI caused the rise [blog].	Counts from logs: jobs, merges, incidents, cost. Narrow, but recorded.
Delivery	Different	Self-serve SaaS and GitHub app, open-source router [site] [blog].	Runs on three Macs for one person (Marton's logs show). Needs a GitHub-native runner before anyone else can use it (inventory).
Size of company	Different	16 people, $17.7M raised (inference), 500+ organisations (claim) [yc] [press] [site].	Two founders, no outside customer.
Price	Different	$0 / $50 per engineer per month / 5% of routed cost [site] [blog].	Not set.
Speed	Same	14 days from install to first executive report (Reducto, claim); enterprise go-live under 2 weeks (claim) [site].	Not measured with an outside team.

6. Wedge overlap

Wedge	Fit	Why
W2 Seeing whether your AI agents actually produce	direct	Same question for the same buyer: what output did the agents give, what did it cost. They ship AI Impact, Token Intelligence and an Agent Observability section [site]. They are funded and have named customers [press] [site].
W1 Trusted 24/7 coding for small startup teams	adjacent	Same buyer, and they sit in front of Claude Code, Codex and Cursor [blog]. They do not run or merge the agents' work, so they are a possible partner as much as a rival (inference).

7. Whitespace (where they stop)

Layer gap: They measure and route; they do not run the agents or land the work [site] [blog]. To own it we need one outside team whose agent fleet our system runs for 4+ weeks, with merged work counted.
Number gap: Their headline number is modelled output, with the caveat that it is not recorded hours or proof of cause [blog]. They do not publicly count work lost (jobs that silently stopped, changes that never merged) (inference: not seen on site or blog; UNVERIFIED that none exists). To own it we need a before/after count from a real fleet, for example share of agent jobs that reached the main code.
Buyer, price and speed gaps: not found. They have a free tier, a $50 seat price and a 14-day claim [site], so we cannot win on cheaper or faster.

8. For the YC application

Proven today: Marton's logs show his system runs and checks the work itself: about 5,600 jobs, about 160 failure records that each added an automatic check, and a heartbeat that reports what is stuck and what silently stopped. Weave's public numbers are modelled estimates read from PRs [blog]. The same logs also show weak spots: about 15% of jobs reached the main code, and one launch used 39% of a week's AI budget in 14 hours.
Draft answer to "how are you different from Weave?": DRAFT. "Weave measures the output of coding agents from the outside. Our system runs a fleet of them, and in our own logs we see what got stuck, what failed silently and what it cost." Founders say it their own way.
Future claim (needs proof): "We show a team how much of their agents' work actually landed, and fix the leaks." Needs one outside team's agent fleet run for 4+ weeks, with a before/after count of work that reached main.
What we understand that they don't: the real problem with agents is work that is lost or silently stops, not how fast PRs arrive. needs proof on an outside team; Marton's logs show it on his own fleet (about 160 failure records).
Copyability (M3): easy. They already have Agent Observability, the Router escalation classifier and 50M+ prompts analysed (claim) [site]. Adding silent-failure and lost-work counts looks like a feature, not a new business (inference). Going further to run the work would pull them off a neutral measurement position. They have sales staff and GitHub access already, so the measurement edge is the easy part to copy. M3 cannot score above 3 on W2 against Weave.
Why now (M4): None found. Their Q3 report shows AI spend per employee up 2.5x between June and September 2026 (they claim; they also say some cost types were first recorded in Q3, which overstates the growth) [blog], but that is self-reported demand, and it helps Weave at least as much as us.
Do not claim: that we beat them on measurement or have a bigger data set. That anyone but Marton has run the system. That we are the first to track agent cost and failures. Anything about the Router's quality. That our "Jev" is related to the model in their blog post (UNVERIFIED, see section 9).

9. Ideas worth taking

Give the buyer one weekly number a boss can read, with a baseline: "output per engineer vs your own last quarter" [site]. Same shape as a before/after count.
Session state beats prompt text for routing decisions: they report accuracy rising from 0.28 to 0.61 when the classifier saw prior tool calls and errors [blog]. The inventory lists a typed-probability model called Jev in Marton's OS; Weave's post is about a model of that name. Whether they are the same thing is UNVERIFIED, so check before using this.
A free tier plus an open-source piece that drives installs (router repo, 5.6k stars, they claim) [site]. A way to get a first outside team without a sales call.

10. Unknowns

Real revenue, customer count by size and renewal rate. Not disclosed. Look at later press or Business Insider coverage of the July 2026 round.
Whether the WorkWeave to Weave rename had a date and a reason. No post found; ask the founders in public talks.
What the Agent Observability section does in detail (tried a guessed page address and got 404). Open the Code Intelligence and AI Impact pages from the site menu.
Whether UK or European companies buy. Only US customers and San Francisco roles seen.
The Field CTO and forward-deployed manager job ads (403 on fetch). They would show deal size and sales model.

Sources

Weave home [site]
Weave pricing [site]
Weave customers [site]
Weave enterprise [site]
Weave blog index, newest posts 16 Oct 2026 [blog]
Weave Router 2.0: Astra quality at half the cost, 9 Sep 2026 [blog]
Engineering output rose 87% in Q3 as AI spend grew 2.5x, Oct 2026 [blog]
Why you shouldn't use Jev for coding agents and routing, 23 Sep 2026 [blog]
YC company page, including open roles [yc] [jobs]
Weave raises $13.5M Series A, 28 Jul 2026 [press]
Weave Launch: AI to quantify engineering work (Fondo repost), dated 11 Feb 2026 [press]
Dealroom: Weave raises $13.5M (from search results; page not opened) [press]
Search result: WorkWeave job listings (page blocked, title only) [jobs]

Weave

What they do

How close to us

Wedges

Who buys

Full teardown