7k+
Daily active users
0 → 7k in 4 months · +30% WoW
11.8k
GitHub stars
70+ contributors
2
Enterprise contracts
Closed during the batch
In use at
Open-source · used by engineers across the world.
Thesis
Superset is the agentic IDE for the engineers already living in the future: tokenmaxxers running 10+ coding agents in parallel.[1][2] Today, it removes the pain of managing worktrees, terminal tabs, and tmux sessions. But the bigger vision is the control plane for software factories.[5]
As coding agents become developers, engineers become what Sequoia's Alfred Lin calls fleet commanders[13]: spinning up agents to refactor code, write tests, build UI, fix bugs, and open PRs in parallel. Over time, the job shifts from writing every line of code to designing, supervising, and improving the systems that write code.[5]
Superset is building the workspace for that future: a place to dispatch, isolate, monitor, review, and merge work from hundreds of agents.
If coding agents are the workers, Superset is the factory floor.
- 01
The bottleneck moved from writing code to orchestrating agents. Ramp's internal agent ships 30% of merged PRs.[7] Stripe's Minions produce 1,000+ PRs/week.[8] Spotify's Honk writes ~50% of their PRs.[9] Every major eng org is reinventing this layer. Superset is the off-the-shelf version.
- 02
Open-source + CLI-agnostic is strategic. Open-source is the wedge — ICs adopt the IDE in an hour, the community ships extensions, enterprise security can read the source. CLI-agnostic is the bet on where frontier labs are actually investing: the agent harness lives inside their CLIs, not their public APIs, because that's where margins and developer mindshare both compound. Superset runs each CLI as a subprocess against the user's existing subscription — structurally aligned with where new features will keep landing.[17]
- 03
The vertical integrators and agent teammates are collapsing into each other — Superset sits a layer above both. The vertical integrators (Claude Code, Codex, Antigravity) and the agent teammates (Devin) are racing toward the same shape — own the model and the harness — absorbing the first wave of agent IDEs along the way: Cursor trained Composer and joined the integrators; Cognition swallowed Windsurf. Both camps end up closed, single-vendor stacks. Superset plays the third layer above them: the orchestration layer — harness-agnostic, open-source, composable — the only position that gets stronger as the stacks below it close.
- 04
Three repeat founders, all ex-CTOs at YC companies. Kiet (Onlook W25), Satya (Untether Labs W23), Avi (Adam W25). They shipped the MVP at a YC hackathon, won, and grew to 7k+ DAUs (+30% week-over-week), 11.8k GitHub stars, and two closed enterprise contracts inside six months.[6]
Software development is becoming fully autonomous
Problem
The first wave of AI IDEs was built for one engineer, one agent.
Cursor, GitHub Copilot, Windsurf, and the rest were all designed around a single human paired with a single agent — one chat window, one editor, one diff at a time. That shape was necessary at the time: the underlying coding models still needed close supervision, manual edits, and frequent human overwrites for anything non-trivial. Which is exactly why these products were all built as forks of VS Code with the text editor as the primary surface and the agent as an assistant in the sidebar.
Then the models got good. Claude Sonnet 4.6 / Opus 4.7 (1M context), GPT-5 / Codex, and Antigravity can now write whole features, review their own PRs, run their own tests, and ship without a hand-holder. The 5–10× productivity gain everyone talks about only shows up when you run several of them at once — and the new top 5–10% of builders, the tokenmaxxers, have already shifted: instead of waiting idly while one agent works, they spin up many in parallel and review the outputs as they finish.
But running several at once is where the friction lives. You're juggling terminal tabs, learning tmux, setting up git worktrees by hand, switching branches, re-orienting yourself every time you come back to the agent you started twenty minutes ago, then reviewing diffs across scattered sessions. The VS-Code-fork architecture wasn't designed for this — it was designed for one human typing.
Most engineers respond by either (a) running one agent at a time and idling while it works, or (b) building bespoke internal infrastructure. The Fortune 500 dev orgs took option (b): Ramp built Inspect,[7] Stripe built Minions,[8] Spotify built Honk.[9] OpenAI coined the term harness engineering to describe the practice — and used it internally to ship a beta product of roughly 1M lines of code with zero manually-written source code.[4] Each of them is now publishing engineering posts about how essential this layer is — and how expensive it was to build.
5–10×
Code output gain
With parallel coding agents
6+ mo
In-house build time
Per Fortune 500 engineering case study
1
Click to a worktree
cmd+N in Superset
Why Now
Tokenmaxxer to Software Factory.
Andrej Karpathy has publicly described the exact product Superset is building — a proper "agent command center" IDE for teams of coding agents.[16] Alfred Lin (Sequoia) calls these builders fleet commanders and reports they're already 3–5× more productive than the median.[13] In the same six-month window, Ramp, Stripe, Spotify, and OpenAI all publicly described internal background-agent infrastructure.
The arc
Tomorrow · the team
The software factory
The org-scale version: dozens to hundreds of agents triggered upstream from Linear, GitHub, and Slack — running in sandboxed cloud workspaces, on schedules, and on event triggers. Humans move off the factory line: they architect the system and review the edge cases, but don't write or review any diff.[5] Ramp Inspect, Stripe Minions, and Spotify Honk are early implementations.[7][8][9]
The arc is the thesis. The same primitives — parallel agents, isolated worktrees, diff review, event-triggered orchestration — show up in both the tokenmaxxer's IDE today and the team's factory tomorrow. Superset's wedge is the open-source IDE an IC adopts inside an hour; the durable economic unit is the team-platform layer (Remote Workspaces, Automations, Shared Memory) that the same engineer pulls in once their team starts running parallel agents at scale. Every tokenmaxxer is a future buyer of the factory; every factory was started by a tokenmaxxer.
tmux grids are awesome, but i feel a need to have a proper "agent command center" IDE for teams of them, which I could maximize per monitor. E.g. I want to see/hide toggle them, see if any are idle, pop open related tools (e.g. terminal), stats (usage), etc.
Andrej Karpathy[16]
The best builders today do not try to write every line of code. They use their taste and judgment to determine what to build and plan how to build it. Then they spin up parallel agents: one refactoring, one writing tests, one working on the UI, all operating with clear intent.
Alfred Lin[13]
We frequently see engineers spinning up multiple minions in parallel, to enable them to parallelize the completion of many different tasks. A typical minion run starts in a Slack message and ends in a pull request which passes CI and is ready for human review, with no interaction in between.
Stripe Engineering[8]
There's no limit to how many sessions you can have running concurrently, and your laptop doesn't need to be involved at all. When background agents are fast, they're strictly better than local: same intelligence, more power, and unlimited concurrency.
Ramp Engineering[7]
The top 5% to 10% of builders are now 3 to 5 times more productive than they were a year ago. The median builder is up maybe 10% to 20%.
Four things converged inside a single six-month window.
Model reliability crossed the bar. The inflection point was December 2025: Claude Opus 4.7 (1M context) shipped, GPT-5/Codex matured, and Anthropic and OpenAI's coding harnesses leveled up around them — Skills, subagents, slash commands, hooks, persistent memory. The experimental pattern of one human running many agents in parallel tipped into a production-grade workflow within a quarter.
Model-provider pricing punished the API-wrappers. In a six-week stretch this spring, Anthropic cut third-party harnesses off from Claude subscriptions, then reinstated access via a capped Agent SDK credit pool billed at full API rates — a step-change cost increase for any tool that wraps the API.[17] Superset's CLI-agnostic architecture sidesteps the squeeze: agents run as subprocesses against the user's existing Claude Code, Codex, or Antigravity subscription. The same structural insulation would absorb the next pricing shift, whichever provider triggers it.
The flagship orgs went public. Ramp,[7] Stripe,[8] and Spotify[9] have all disclosed in-house builds of the software-factory model in the last six months — making the architecture legible to enterprise buyers and putting Superset in position to be the off-the-shelf fleet console.
The denominator flipped. AI now writes 42% of committed code per Sonar's 2026 survey of 1,100+ professional developers — up from Google's >25% disclosure in October 2024 and Microsoft's 20–30% in April 2025 — and developers themselves expect 65% by 2027.[23][24][25] At the frontier, 89% of code committed at Cognition is written by Devin, up from 13% in December 2025.[22] When most code is agent-written, the scarce input is no longer the model — it's the dispatch, isolation, and review layer the human operates. That layer is the product Superset sells.
The majority of code will be agent-written within two years
Chart
Dated public datapoints: Google reported >25% of new code AI-generated (Oct 2024); Microsoft 20–30% of repo code (Apr 2025, upper bound shown); Sonar's 2026 survey of 1,100+ professional developers measures 42% of committed code today and a 65% developer expectation by 2027. The leading edge is further out — Cognition reports 89% of its own committed code is written by Devin (May 2026), up from 13% in Dec 2025.
Source · Fortune (Oct 2024) · CNBC (Apr 2025) · Sonar State of Code 2026 · TechCrunch (May 2026)
The software development life cycle must change. To get the most out of coding agents, they need to shift not only to the background, but also to the cloud, on schedules and on event triggers, to automate the new bottlenecks springing up in the lifecycle.
How It Works
From local worktrees to the cloud command center.
v1 — the IDE for parallel agents running locally.
The MVP that came out of the YC hackathon: a desktop IDE built around cmd+N spawning isolated git worktrees, with any CLI coding agent (Claude Code, Codex CLI, OpenCode, Antigravity CLI) running inside each one. Notifications when each agent finishes. GitHub-style diff review across sessions. One-click open in VS Code, Cursor, JetBrains, or Xcode. Linear integration to fire agents at backlog tickets directly.
v1 found product-market fit fast — 7k+ DAUs growing 30% w/w, 11.8k GitHub stars, front-page HN in five months.[14] But worktrees don't scale: shared CPU, RAM, and disk space across many parallel agents, overlapping builds, messy envs and secrets, all bound to one machine. The worktree model worked beautifully — until it ran into the physical limits of the laptop it was running on.
v2 — the platform layer that unbinds Superset from the laptop.
The pivot: move the workloads to the cloud. v2 shipped in early May 2026, layering the missing infrastructure underneath the IDE — and notably, the team kept the v1 IDE shipping, the Discord swarming, the DAUs compounding 30% w/w, and the enterprise contracts closing the entire time they were building it.
Remote Workspaces. Ship a session to the cloud; continue working from anywhere; share with teammates. Solves the laptop-bound limitation that internal background agents at Ramp and Stripe both name as their core unlock.[7][8] Start on your laptop, continue on your phone, scale to as many agents as you need.
Automations. Schedule agents to pick up tasks for you later — or fire them on events: a new issue opened in Linear or GitHub, a Sentry error detected, a failed CI run, an inbound Slack message.
Superset CLI. Gives agents new superpowers and unlocks programmatic workflows — turning Superset itself into a primitive other agents can call.
MCP refresh. Works with v2 + an expanded toolset. Integrates across the apps where product and engineering work happens — Slack, Linear, GitHub, and more. Message @Superset in Slack to kick off a new task without ever opening the IDE.
Mobile · coming soon. Talk to your codebase from anywhere. Kick off a task on your phone walking to a meeting, review the diff on the train, ship from the couch. Same Remote Workspace, different surface.
Open-source, with the GitHub trajectory of a category leader
TractionFive months from launch to 7k+ daily active users growing 30% week-over-week, 11.8k GitHub stars (70+ contributors), and front-page HN on launch day.[14] Growing by word-of-mouth inside world-class engineering orgs — Wix, Netflix, Microsoft, DoorDash, Vercel, Stripe, Datadog, ByteDance, NVIDIA, OpenAI, Salesforce, Cloudflare, and Atlassian. Two enterprise contracts closed during the batch. The May 2026 Launch HN for v2 surfaced the demand curve directly: users reporting 40–50 concurrent agent sessions across several repos, switchers from Conductor, and engineers who had moved their entire daily workflow into Superset since December.[26]
The Long Arc
The IDE is the wedge. The Software Factory is the prize.
A Software Factory is an org-scale operating model where parallel coding agents take work from anywhere it originates, run in isolated environments, and return finished work for human review. Engineers don't hand-write or hand-review every diff; they architect the system, calibrate the loops, and decide which edge cases get attention. Today's tokenmaxxer is already running a small version of this on their laptop. Tomorrow's team runs a big one. Superset's IDE is the wedge into both.
Every major eng org is converging on the same architecture.
The loop. Take a task. Spawn a sandboxed agent with the right tools and codebase context. Write code. Run CI. Open a PR. Review and iterate. Repeat in parallel — dozens, hundreds, thousands of times. The human's job has narrowed to architecting the loop and reviewing what comes out the other end, not typing the diff. Ramp's Inspect, Stripe's Minions, and Spotify's Honk are three implementations of the same loop.[7][8][9]
The pillars. Three primitives make the loop work: isolated compute (each agent's blast radius is sandboxed), event routing (work flows in and out automatically), and governance (humans review, approve, and shape what merges).[3] Superset's v2 maps cleanly to all three. Together they're what moves humans from on the line to on the loop.
The economic argument. Every team that has done this in-house has spent 6+ months and senior engineers to build it. Most companies will not. Superset's bet: nobody actually wants to build their own factory machinery — they just want the IDE that ships with it, so their engineers can step off the line on day one.
How do you build a 100+ agent software factory?
Control plane
Directs, governs, and optimizes work
Execution plane
Builds and runs work at scale in secure envs
Learning plane
Compounds agent performance
Delivery plane
Turns jobs into completed outcomes
There's no limit to how many sessions you can have running concurrently, and your laptop doesn't need to be involved at all. When background agents are fast, they're strictly better than local: same intelligence, more power, and unlimited concurrency.
Market
Three concentric markets. Superset has commercial validation in all three already.
Near term — tokenmaxxers. Individual engineers paying $200+/month for Claude Code or Codex who would happily run more agents if the friction were lower. 7k+ DAUs growing 30% week-over-week, 11.8k stars, and front-page HN inside five months — the natural-demand signal.[6][14] Monetization is live via a $20/seat Pro tier, with paid seats compounding weekly. Bottom-up, the wedge market alone is $60B — 30M professional developers × ~$2k/year of agentic tooling spend.
Mid term — agentic coding teams. Small-to-mid engineering orgs (5–200 engineers) standardizing on a parallel-agent workflow as a team. Two enterprise contracts closed during the batch. The revenue stack is per-seat IDE plus the org-level platform features — Remote Workspaces, Automations, shared memory, admin controls, SSO/RBAC, audit logs — that get pulled in once an IC's workflow becomes a team workflow.
Long term — Software Factories. Dozens to hundreds of agents triggered upstream from Linear, GitHub, Slack, and on-call — running in sandboxed cloud workspaces and shipping PRs continuously. Ramp Inspect, Stripe Minions, and Spotify Honk are early in-house implementations.[7][8][9] Every team that wants this without spending six months building it is a future Superset customer.
The capital markets have already re-rated the category. Cursor crossed $2B ARR in February 2026 — the fastest zero-to-$2B in B2B software history — and is raising $2B+ at a $50B pre-money valuation, forecasting a $6B+ run-rate by year-end.[19] Claude Code passed a $2.5B run-rate the same month, more than doubling in six weeks.[20] Codex crossed 5M weekly users by June 2026, up from ~600k in January.[21] Cognition raised $1B at a $26B post-money on $492M ARR.[22] The read-through for Superset is not that it competes for this revenue — it's that every one of those subscriptions is an agent Superset's users already pay for. Superset sits on top of the fastest-growing spend pool in software without carrying the model margin or the inference bill.
The agent suppliers' revenue went vertical inside 14 months
Chart
ARR run-rate milestones as publicly reported. Cursor: $100M (Jan '25) → $2B (Feb '26), forecasting $6B+ by end of 2026. Claude Code: ~$1B within six months of GA → $2.5B+ (Feb '26). Cognition (Devin): $37M (May '25) → $492M (May '26). Every line on this chart is a subscription Superset orchestrates rather than competes with.
Source · TechCrunch (Apr 2026) · Constellation Research (Feb 2026) · TechCrunch (May 2026)
A typical minion run starts in a Slack message and ends in a pull request which passes CI and is ready for human review, with no interaction in between. Over 1,000 PRs per week are completely minion-produced.
Competitive landscape
Two camps are collapsing into each other. Superset plays the layer above them.
The vertical integrators (Claude Code, Codex, Antigravity) and the agent teammates (Cognition) are converging on the same race — own the model and the harness — absorbing the first wave of agent IDEs along the way: Cursor trained Composer and joined the integrators; Cognition bought Windsurf. Both camps end up closed, single-vendor stacks. The third position — the orchestration layer that sits above them all — only gets stronger as they collapse, and Superset is the only company there combining category-leading OSS traction (11.8k stars, 7k+ DAUs growing 30%/wk), a harness-agnostic architecture aligned with how frontier labs ship features, a shipped v2 (Remote Workspaces, Automations, Shared Memory, Chat 2.0), and three repeat ex-CTO founders running the whole loop in public.
Superset's positioning
As the stacks collapse into single-vendor verticals, Superset is the neutral orchestration layer above them — harness-agnostic, open-source, composable. Claude Code, Codex, Cursor, OpenCode, and Antigravity become the workers; Superset is the one place to dispatch, isolate, observe, review, and merge their work — and the only player whose position improves as every vendor races to own its own model and harness.
Founder deep dive
The founding moment — and why it had to be these three.
Founders
Kiet Ho
Repeat FounderCo-founder & CEO
Repeat founder. Previously co-founder/CTO at Onlook (YC W25). Earlier at Amazon and ServiceNow. Open-source maintainer; conceived Superset at a YC hackathon after running into the friction of managing many Claude Code agents at once.
Satya Patel
Repeat FounderCo-founder
Former CTO at Untether Labs (YC W23). Prior engineering at Scribe, Google, Amazon, and Facebook. Systems-oriented; authored Superset's worktree isolation and persistent daemon architecture.
Risks & mitigations
What we're watching
References
- [1]Superset — YC Profile
- [2]Superset — Company Website
- [3]background-agents.com — The Self-Driving Codebase
- [4]OpenAI — Harness Engineering: Leveraging Codex in an Agent-First World
- [5]Simon Willison — Software Factory
- [6]Superset on GitHub
- [7]Ramp Builders — Why We Built Our Background Agent (Inspect)
- [8]Stripe Engineering — Minions: Stripe's One-Shot End-to-End Coding Agents
- [9]Spotify Engineering — Honk: Spotify's Background Coding Agent (Part 1)
- [10]Cursor — Towards Self-Driving Codebases
- [11]Superset 2.0 launch on Product Hunt
- [12]Superset launch on Bookface (YC internal, P26)
- [13]Alfred Lin (Sequoia) — Developer to Fleet Commander
- [14]Show HN — Superset: terminal to run 10 parallel coding agents
- [15]How Superset built the IDE for AI agents on Vercel
- [16]Andrej Karpathy — "agent command center" IDE for teams of agents (X)
- [17]VentureBeat — Anthropic reinstates OpenClaw and third-party agent usage on Claude subscriptions — with a catch (May 13, 2026)
- [18]Google Developers Blog — Transitioning Gemini CLI to Antigravity CLI (May 19, 2026)
- [19]TechCrunch — Cursor in talks to raise $2B+ at $50B valuation as enterprise growth surges (Apr 17, 2026)
- [20]Constellation Research — Anthropic's Claude Code revenue doubled since Jan. 1 ($2.5B+ run-rate, Feb 2026)
- [21]Constellation Research — OpenAI touts broadening Codex usage with 5 million weekly active users (Jun 2026)
- [22]TechCrunch — AI coding startup Cognition raises $1B at $25B pre-money valuation (May 27, 2026)
- [23]Sonar — State of Code Developer Survey report 2026 (42% of committed code is AI-written; 65% expected by 2027)
- [24]Fortune — Over 25% of Google's code is written by AI, Sundar Pichai says (Oct 2024)
- [25]CNBC — Satya Nadella says as much as 30% of Microsoft code is written by AI (Apr 2025)
- [26]Launch HN — Superset (YC P26): IDE for the agents era (May 2026)
