Replicas
Cloud platform for running coding agents like Claude Code or Codex remotely.
On the record
Checkpoint: OpenCode support ships inside Replicas, testing whether multi-harness demand is real or teams stay on one vendor
The call
Replicas sits directly in a lane that Anthropic, OpenAI, and GitHub are already building into their own products, leaving a two-person team to compete on flexibility alone.
Replicas lets developers run AI agents in the background and trigger them from Slack or GitHub, without having to pick one AI model. The problem: GitHub Copilot, Claude, OpenAI, and Cursor have already built their own versions of this and put them directly inside their own products, which they control completely and give away for free. Since Replicas doesn't own the code it touches, doesn't lock developers in, and doesn't have special data or regulations protecting it, the four big platforms can copy everything Replicas does without paying Replicas anything. The one thing that could save Replicas is if developers strongly prefer working with a neutral middleman instead of being locked into one company's AI agent system, but so far they have not shown that preference.
Why
01
AI pressure: Anthropic Claude Code, OpenAI Codex, Cursor Background Agents. These labs shipped their own cloud and background execution in Feb 2026, removing the need to route through a separate layer like Replicas. Could AI do the core job? Yes, likely.
02
Bigger players: GitHub Copilot (Microsoft). Copilot now runs Claude Code and Codex inside its own plan, covering Replicas' core job with 10M+ developers already on the platform. Could a giant copy it? Yes, easily.
03
What protects it. Multi-harness flexibility for teams that don't want to lock into one vendor's agent, though this is easy for any of the above to copy.
What would change this call
GitHub Copilot ships native seat-based cloud agent execution inside its core plan. Would commoditize Replicas' multi-harness orchestration and remove its main pitch.
Cursor's reported xAI/Colossus infrastructure deal and $60B acquisition option is confirmed executed. Would show a rival consolidating infrastructure scale Replicas cannot match.
Replicas' named customers (Mintlify, Composio, Helicone, Knowunity) publicly churn or switch to Copilot/Cursor. Would show its claimed usage and PR-share numbers are not sticky.
Cursor's reported xAI/Colossus infrastructure deal and $60B acquisition option is confirmed executed. Would show a rival consolidating infrastructure scale Replicas cannot match.
Replicas' named customers (Mintlify, Composio, Helicone, Knowunity) publicly churn or switch to Copilot/Cursor. Would show its claimed usage and PR-share numbers are not sticky.
Key numbers
- Used by 20+ YC startups
- Team size of 2 people
- Pricing runs $120 to $300 per seat monthly
Named risks
- No structural moat: core orchestration function is being absorbed natively by better-funded model vendors and GitHub.
- Extremely small team (1-2 people historically) creates key-person and support-capacity risk.
- Funding amount and revenue remain undisclosed/unverified beyond standard YC participation.
- Category described by third-party analyst as crowded and consolidating fast, with well-capitalized rivals (Devin, Cursor) accelerating.
- Traction metrics (20+ customers, 30% of PRs, 95% dogfooding) are vendor-claimed and unverified by independent third parties.
Rivals, as read in July 2026
| Rival | Direction | Evidence |
|---|---|---|
| GitHub Copilot (Microsoft) | losing to | Offers third-party agents (Claude Code, Codex) inside its own plan, owns 10M+ developer PR surface, 2026 |
| Cursor (Anysphere) | losing to | Shipped background agents Feb 2026 and reportedly secured $60B acquisition option/Colossus infra deal, May 2026 |
| Cognition's Devin | losing to | Raised ~$1B at ~$26B valuation around June 2026, signaling category consolidation |
| OpenAI Codex | losing to | Reached 2M+ weekly active users by March 2026, up from ~5% to ~40% of Claude Code usage since Sept 2025 |
| Anthropic Claude Code | losing to | Introduced native 'Agent Teams' cloud execution feature in Feb 2026 |
| Ara, Ellipsis, boxes.dev | contesting | Direct peers offering near-identical cloud/background coding-agent orchestration as of mid-2026 |
Questions people ask
- Is Replicas profitable in 2026?
- No revenue or profit figures have been disclosed. Replicas is a YC Spring 2026 (P26) startup with seat-based pricing ($120-$300/seat/month per Ry Walker Research, 2026-06-10) but no confirmed ARR or funding amount beyond standard YC participation.
- Could AI replace Replicas' own product?
- Yes, plausibly. Replicas' core function, cloud sandbox orchestration for coding agents, is being absorbed natively by Anthropic, OpenAI, GitHub, and Cursor, all of which shipped competing background-agent features in February 2026, before Replicas launched publicly.
- Who is Replicas' biggest competitor?
- GitHub Copilot is the most structurally threatening rival: it already offers third-party agents like Claude Code and Codex inside its own plan (per eesel.ai, 2026-07-01) and owns the code/PR surface plus distribution to over 10 million developers.
- Who owns Replicas?
- Replicas is privately held and venture-backed, part of Y Combinator's Spring 2026 (P26) batch, founded by CEO Connor Loi (ex-Ramp, Shopify, IBM) and a co-founder. No acquisition or ownership change has occurred as of the July 2026 Product Hunt launch.
- Does Replicas have a defensible moat?
- No structural moat has been identified. It has no proprietary data, regulatory cover, lock-in, or owned distribution; its multi-harness orchestration wrapper sits directly in the shipping lane of Anthropic, OpenAI, GitHub, and Cursor, which already ship equivalent cloud-agent features natively.
- How big is Replicas' customer base?
- Replicas claims (vendor-claimed, tryreplicas.com) to be used by 20+ YC startups and that customer engineering teams collectively ship over 30% of pull requests through the platform; named customers include Mintlify, Composio, Helicone, and Knowunity.
Timeline
- 2026-07-26Replicas launches on Product Hunt as 'Run Claude Code and Codex in the cloud'
- 2026-06-10Ry Walker Research flags solo-founder risk and crowded, fast-consolidating category
- 2026-06-07Rival Cognition's Devin reportedly raises ~$1B at ~$26B valuation
- 2026-06-02Replicas' BYO-model-key business model (customers bring own API keys) described publicly
- 2026-05-12Rival Cursor (Anysphere) reportedly gets Colossus access and $60B acquisition option via xAI/SpaceX deal
- 2026-04-07Industry comparison notes Cursor, Claude Code, Codex, Windsurf all shipped cloud/background agents in Feb 2026
- 2026-04-04YC jobs listing shows Replicas team size of 2, hiring a Founding Engineer
- 2026-04-01Replicas launches via Launch YC as 'End-to-End Background Coding Agents'
- 2026-02Cursor, Claude Code, OpenAI Codex, Windsurf all ship background/cloud agent features within a two-week window
Revision history
27 July 2026
First assessed. Little protection left. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources
- Comprehensive Comparison & Rankings (2026-04-07)
- OpenAI Codex (AI agent)
- The 8 best Devin Fusion alternatives in 2026 (2026-07-01)
- 8 Cloud Coding Agents That Open PRs While You Sleep (2026-06-07)
- Best AI Coding Agents in 2026 (2026-05-12)
- Replicas: Run any coding agent harness in the cloud
- Background Coding Agents for Engineering Teams
- Founding Engineer at Replicas (2026-04-04)
- Ry Walker Research (2026-06-10)
- Replicas: Run Claude Code and Codex in the cloud (2026-07-26)
- Replicas: Run Claude Code and Codex in the cloud (2026-06-02)
- Ara: The Cloud Coding Agent