Conductor (YC Summer 2024): teardown vs XMarts

In one line: Conductor is an app where an engineer runs several AI coding agents at once, each in its own copy of the code, then reviews and merges their work [site] [docs]. Threat: 3 because it has the same buyer and problem as W1, is funded ($22M Series A, Mar 2026), is live, and names big-company users [blog] [yc]. It is a control room for agents, not a trusted overnight builder, so our answer must be about trust and learning, not about running agents in parallel.

1. Fact file

2. Their proposition, reverse engineered

3. Positioning

• Category they claim: "first AI orchestrator" and the human interface to AI organizations [jobs]. The site calls it the best way to run multiple coding agents in parallel [site].
• Category a YC partner would put them in: developer tooling for coding agents, a control plane or workbench over Claude Code and Codex (inference).
• What they want to be known for: the place where teams assign and supervise coding agents, with model vendors supplying the intelligence [press].
• Where they sit on the map: L1/L2/L3 = 1/3/1, buyer = enterprise by named customers, developer-led adoption.

4. Three layers vs ours

5. Same and different

6. Wedge overlap

7. Whitespace (where they stop)

• Layer gap: no company brain and no learning from results [docs] (inference). To own it we need a rule-change log tied to checked runs, with rework falling over weeks. This is the W1 plus W8 angle in the brief.
• Buyer gap: they build for engineers. Nothing is aimed at a finance or operations leader who wants a result owned [site]. Also unknown in the UK: SF office only on the job ad [jobs]. To own it we need a non-engineer buyer or a UK team saying they have the pain.
• Number gap: public proof is logos and growth, with no hours saved, bugs avoided or revenue [blog]. To own it we need one workflow with a before/after number (merged changes per week, rework rate, hours).
• Price gap: not a gap. $50 and $60 a seat is cheap, and the base is free [site].
• Speed gap: not a gap. Download and run [site].

8. For the YC application

• Proven today: Marton's logs show his system runs agents from three makers around the clock: about 5,600 jobs, each built in its own copy of the code and checked by a different AI, then automatic tests. About 160 failure records each added an automatic check. This is on his own business only, and about 15% of jobs reached the main code.
• Draft answer to "how are you different from Conductor?": DRAFT. "Conductor is a window where a person runs and reviews coding agents. Our system already runs a queue of agents day and night, has a different AI check every job, and turns each failure into an automatic check, on our own business." Founders say it their own way.
• Future claim (needs proof): "Our agent team gets better with every accepted or rejected change, so you review less over time." Needs one outside team using the system for 4+ weeks, with a rework rate or review time that falls.
• What we understand that they don't: the limit on agent work is trust in what comes back, not the number of agents you can run. needs proof; Marton's logs show it on his own system (failures are named and each gets a check), but no outside team has used it.
• Copyability (M3): easy. They already ship code review, routines (automation), an API, an MCP server and cloud workspaces [changelog]. An automatic review step is a feature for them, and it fits their engineer buyer. They have $22M and an 8-person team that ships regularly (two releases in the first half of Oct 2026: 0.90.0 on 2 Oct and 0.91.0 on 15 Oct) [blog] [changelog]. A learning loop across runs is harder (medium), but nothing blocks them. Not hard, so M3 can not score above 3 against Conductor.
• Why now (M4): None found in this teardown. Do not use "agents run for hours now": that is general AI progress. (Their own risk, not our edge: Anthropic announced subscription changes for third-party Agent SDK tools on 13 May 2026 and delayed them on 15 Jun 2026 [blog].)
• Do not claim: that our system is "trusted" or "24/7 reliable" (landing was unreliable in October; Marton's logs show 15% of jobs merged). That anyone but Marton has used it. That Conductor lacks review, cloud or API (it has all three). "Parallel agents" as our headline (it is their product).

9. Ideas worth taking

1. A mobile app to review and merge, because Slack is "too noisy" for agent work [press]. A leader can approve from a phone.
2. Hand over the session, not just the diff, so a teammate keeps the context [press].
3. An API so other tools and agents can start work, which makes the product a base others build on [press].

10. Unknowns

• Revenue, paying seats and how many of the claimed logos are paid teams. Look for a later funding post, a podcast or a talk.
• What the enterprise page really offers; the fetched text showed no features or logos [site]. Check the rendered page.
• Whether the renames (BookMe, Melty, Chorus) are real and how they link to Conductor. Ask the founders or check old launch posts.
• UK or EU customers. Look at the customer logos and any careers page outside the SF job ads.
• Whether "routines" (automation, v0.85.0) can chain work or learn from results. Read the docs page for routines.
• Whether the "checks" doc page measures anything beyond CI status.

Sources

• conductor.build [site]
• Conductor pricing [site]
• Conductor Enterprise [site]
• Conductor docs introduction [docs]
• Conductor changelog, newest release 0.91.0 on 15 Oct 2026 [changelog]
• Conductor's Series A, 30 Mar 2026 [blog]
• cmd joins Conductor, 2 Apr 2026 [blog]
• Claude subscription update for Conductor, 15 Jun 2026 [blog]
• Conductor blog index [blog]
• YC company page [yc]
• YC job ad: Product Engineer [jobs]
• Conductor launches multiplayer cloud workspaces, 30 Jul 2026 [press]
• Conductor launches an iPhone app, 2 Oct 2026 [press]
• Multiplayer Coding Agents in the Cloud, podcast with Charlie Holtz [press]
• YC company list data (former names, team size) [yc]