Wato (YC Spring 2026): teardown vs XMarts

In one line: Wato is a shared, permissioned workspace that gives a team's AI agents reviewed memory, shared skills, approved tools and audit trails, reached through one MCP link from Claude Code, Cursor, Codex and similar tools [site] [docs]. Threat: 2, because it targets the same kind of buyer (mid-sized teams running agents) with a live self-serve product, but it is two people with no named customers and its docs show no learning loop yet [yc] [docs].

1. Fact file

Item	Finding	Source
What they say	Site headline: "The shared AI workspace for your company." YC line: "The control point for AI agents at work."	[site] [yc]
Founded, team, funding	YC Spring 2026 (listed as P26), San Francisco, 2 people. YC announced the launch on 3 June 2026. One aggregator says $500k pre-seed on 16 June 2026 (the standard YC deal); Wato has not confirmed it.	[yc] [yc-x] [news]
Founders	Arihan Varanasi (CEO) and Rahul Rejeev (CTO). Both list Stanford CS. No earlier company or product is shown. What YC likely saw: two technical founders on a hot theme (agent infrastructure) (inference).	[yc] [news]
Customers	None named on the site, the launch post, the blog or the YC page. The "Backed by" line on the site shows no names.	[site] [yc] [launch]
Price	Public in the docs: Free $0 (1 member), Personal $40 a month (up to 3 members), Team $200 a month (up to 10 members). The home page gives no price and its FAQ has no answers.	[docs] [site]
Path	No rename or pivot found. Six blog posts, all June 2026. No job ads listed.	[blog] [yc]

2. Their proposition, reverse engineered

Part	What they do
Who buys	Mid-sized firms ("the middle"), not AI-native startups or Fortune 100 [blog]. In practice a team lead or power user who signs up for self-serve, then an admin who sets permissions [docs] (inference).
The pain	Good AI work gets stuck in personal chats, local files and personal accounts, so teammates start over [launch]. Companies want one place for MCP settings, permissions and prompts [launch].
The promise	Work "compounds across a team instead of disappearing into chat threads" [yc].
How it works	Connect Wato once over MCP. Agents in Claude Code, Codex, Cursor and Claude Desktop then get the team's memory, skills and approved tools [site] [launch]. Memory is Markdown pages. Edits are drafts, then a proposal, then approval and merge [docs]. Cloud agent sessions that teammates can view and continue, and automations from schedules or connector events [site] [docs].
The output	Reviewed memory, versioned `SKILL.md` skills, live artifacts and dashboards, and a log of every tool call [site] [docs].
Proof style	None yet. No numbers, customers or case studies anywhere [site] [blog].
Who sells it	Self-serve sign-up with a free plan and card payment, plus "Book a Call" and "Request access" [docs] [site]. No sales hires visible [yc].
Price model	Per team and per member tiers, flat monthly [docs]. Dedicated cloud, customer cloud, VPC and on-premise are listed on the security page, price not shown [docs].
The story	The hard part of enterprise AI is deployment and governance, not model quality [blog]. Agents get reliable when each run is recorded, reviewed and corrected, "not just memory" [blog].

3. Positioning

Category they claim: the "control point" or "collaboration layer" for teams using AI agents [yc] [yc-x].
Category a YC partner would put them in: shared memory plus MCP gateway and governance for agent teams. Close to the Company Brain request, but built around permissions and review (inference).
What they want to be known for: the governed place where a team's agents get trusted context and tools, across whichever AI subscriptions the company already pays for [yc-x] [site].
Where they sit on the map: L1/L2/L3 = 2/2/1, buyer = mid-market, business.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (Marton's logs show)
L1 Company brain	2. Versioned, reviewed team memory that every agent reads, with a Notion sync and a skills library [docs] [site]. Not the whole product, and it is filled by people and agents submitting pages, not learned from watching work (inference from the docs).	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning. Learns from calls and checked work.
L2 Agent orchestration	2. Cloud agents that can be started from MCP, Slack, Teams and Linear, shared sessions teammates can continue, automations, and layered tool permissions [site] [docs]. No claim of several specialist agents handing work to each other.	About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests.
L3 Execution and learning	1. Records every tool call and ties runs to a harness version [docs] [blog]. The blog describes corrections feeding back into memory and rules, but the docs say nothing on run review, run collections or auto-update. So it logs the work; it does not yet check or measure it in the product (inference from the docs).	Every failure gets a record and an automatic check (about 160 incident records since July). Offer-page 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	Same	Agents fail on missing context and weak feedback, not model quality [blog].	Same (brief).
Output files	Same	`SKILL.md` skills and MCP access [docs].	Marton's logs show skills and the vault served through MCP.
Buyer	Similar	Mid-sized firms, team leads, self-serve [blog] [docs].	Target: SMEs and scale-ups (brief). No customer yet.
Who gets the benefit first	Different	Teams that already run Claude Code, Cursor or Codex [site].	We do the work for the buyer, not only supply the shared layer (brief thesis, unproven on a customer).
Review	Different	Humans approve memory and skill changes before agents trust them [docs].	Marton's logs show a different AI checks each job, then automatic tests.
Learning	Different	Described in a blog post, not in the docs [blog] [docs].	Marton's logs show rules added after each failure, with falling feedback lines on one workflow.
Proof	Same (both empty)	No customer or number shown [site].	No customer number yet.
Size and speed	Different	2 founders, live product with free plan [docs].	Two founders, one running system on its own business; nothing sold to a customer.
Price	Different	$0, $40, $200 a month [docs].	Not set yet.
Delivery	Different	Software the customer sets up [docs].	The system plus the two founders (brief).

6. Wedge overlap

Wedge	Fit	Why
W8 The other 90% of an AI skill	direct	Same buyer and same problem area: shared, versioned skills with review before they are trusted [docs]. They also write that corrections should become rules and checks [blog]. The upkeep from real use is argued, not shown as a feature (docs silent). If it ships, this becomes our closest rival.
W2 Seeing whether agents produce	adjacent	Tool-call traces and audit logs, and a post on measuring accepted versus human-corrected outputs [docs] [blog]. They trace access; they do not show what is stuck or silently failed (inference).
W3 Company memory from every call	adjacent	Reviewed team memory with a Gong connector and Notion sync [site] [docs]. Not built for services firms or learning from calls (inference).
W1 Trusted 24/7 coding	adjacent	Cloud agent sessions with a filesystem and git, spawned from Linear or Slack [site]. A shared workspace, not an overnight coding queue with merge checks (inference).

7. Whitespace (where they stop)

Layer gap: they log and govern the work; the docs show no step that checks a result and changes the next run [docs]. To own it we need a rule-change log tied to checked runs, with rework falling, on one customer workflow for 4+ weeks.
Number gap: no customer, no result, no money or time number anywhere [site] [blog]. To own it we need one before/after number a team lead would sign.
Buyer gap, partly: self-serve works for teams that already live in Claude Code or Cursor [site]. A mid-sized firm with no such power users is not served by a tool it must set up itself (inference). To own it we need a UK buyer of that kind saying so on a call.
Price gap and speed gap: not ours. They start at $0 and a sign-up gets a workspace the same day [docs]. We can not win by being cheaper or faster to start.

8. For the YC application

Proven today: Marton's logs show his system runs agents that do work and have it checked by a different AI: about 5,600 jobs, about 160 incident records that each added an automatic check, and offer-page feedback lines down from 44 to 18 as 22 rules were added. Wato's public material shows the memory, the permissions and the audit trail, but no step that checks results [docs].
Draft answer to "how are you different from Wato?": DRAFT. "Wato gives a team's agents shared memory and permissions. Our system also checks the work and turns each failure into a rule, and we run it daily on our own business." Founders say it their own way.
Future claim (needs proof): "Our skills get better from every accepted or rejected piece of work, and we show the number." Needs one customer workflow run for 4+ weeks with a rule-change log and rework falling.
What we understand that they don't: a shared brain is worth the checked results that feed it, not the pages put into it. needs proof on a customer; Marton's logs show it on his own work.
Copyability (M3): easy. They already have traces, versioned skills, run records tied to a harness version and a blog post that names the feedback loop [blog] [docs]. Adding result checks is a feature for them, it fits their buyer (inference). The only brake is that they are two people. Not hard, so M3 can not score above 3 against Wato.
Why now (M4): None found in this teardown.
Do not claim: that nobody ships shared skills for teams (Wato does, with review). "Company brain" or "shared memory" as our edge (they have it). That they lack a feedback loop (their blog describes one, and their docs may be behind their roadmap). That we are better governed than they are (their permission model is more developed than ours for outside teams).

M2 (what YC funds now) is NOT judged here. It comes from the themes of many companies in Pass 3.

9. Ideas worth taking

Memory edits as drafts, then a proposal, then approval, with a diff. It makes a company brain safe to trust [docs].
Tie every run to the exact version of the rules that produced it, so before and after can be compared [blog].
A free plan plus flat monthly tiers by team size. Good for first-week pull from small teams [docs].

10. Unknowns

Whether they have any paying customer or active team. Check their X posts and ask in the YC founder network.
Whether the feedback loop in the blog is built. Read the docs pages on skills, "Common Workflows" and "Reference" (not fetched).
The real funding amount and investors; the "Backed by" line is empty. Look for a YC Demo Day listing.
Whether they sell in the UK. Their pages show no location beyond San Francisco.
Their FAQ answers on "Why is it better than Claude and ChatGPT Teams?" and on-premises. Get them from a demo call.

Sources

watolabs.com [site], page metadata dated 29 Jun 2026
YC company page [yc]
Launch YC: Wato, a shared AI workspace for teams [launch]
Y Combinator on X, launch announcement, 3 Jun 2026 [yc-x]
Wato blog index [blog]
What We Learned Building AI Agents for Real Companies, Jun 2026 [blog]
Why AI Agents Need a Feedback Loop, Not Just Memory, Jun 2026 [blog]
Outcomes versus token spend, Jun 2026 [blog]
Wato docs: Introduction [docs]
Wato docs: Getting Started [docs]
Wato docs: Core Concepts [docs]
Wato docs: Sign Up and plans [docs]
Wato docs: Security and Deployment [docs]
Wato funding profile (aggregator, unconfirmed amount) [news]
Crunchbase: Wato [news]
YC Tier List: Wato (third-party opinion) [news]
Our side: yc/research/system-inventory-2026-10-19.md and brief/xmarts-brief.md

Wato

What they do

How close to us

Wedges

Who buys

Full teardown