AQ vs Hoplite: two ways to run coding agents in the cloud

Published September 8, 2026 · by the AQ team

Choose AQ if you want your team running coding agents together: the real CLIs your engineers already subscribe to, in persistent sessions on a VM you control, with teammates watching and steering the same live terminal from a browser. Choose Hoplite if you want to dispatch tasks from Slack or Linear into vendor-managed cloud sandboxes and get pull requests back, with models served and billed through the platform. Both run coding agents in the cloud so work survives a closed laptop; the difference is the unit of work. Hoplite's unit is the task: a thread spins up a sandbox and the output is a pull request. AQ's unit is the workspace: a durable machine and git worktree where a live session runs until the team is done with it.

Hoplite (hoplite.sh) is a Y Combinator Summer 2026 company that launched publicly in early August 2026. It is moving quickly, so every Hoplite claim below is as of September 2026, verified against hoplite.sh, its documentation and pricing pages, and its YC launch materials.

AQ vs Hoplite at a glance

What is Hoplite?

Hoplite runs autonomous coding agents in isolated cloud sandboxes. You connect a GitHub repository and describe a task, and an agent works it in a real development environment: editing code, running the repository's own tests, booting the app on a managed preview URL, checking its work in a browser, and opening a pull request when done. Four terms carry the product: a project holds a repository, a thread is one task, every thread gets its own sandbox, and the intended output is a pull request.

Two things distinguish it from earlier fire-and-forget cloud agents. First, onboarding imports your local setup: Hoplite transfers local agent sessions and MCP servers and configures the dependencies and CLIs your projects need, so the sandbox behaves like your machine rather than a bare container. Second, dispatch meets you where work lives: threads start manually from Slack, Linear, or iMessage, or automatically from Sentry events, and the draft PR, review loop, and merge can be handled without leaving the thread. Execution is not blind: the agent's actions stream live, and file edits, shell commands, and pull-request actions can each be gated on explicit approval. Models are served through Hoplite's own layer and billed as credits included with each plan.

One caution when researching it: Paxos, the stablecoin infrastructure company, published a July 2026 engineering post titled "Meet Hoplites" about an internal one-shot cloud coding agent it built in-house, reporting that since March 2026 roughly 15 percent of its merged pull requests started with a dispatch to that tool. That is Paxos's own unrelated system, which happens to share the name. The coincidence is still informative: dispatch-a-task, get-a-PR is a pattern serious engineering organizations now run in production, and Hoplite is a way to buy it instead of building it.

What is AQ?

AQ is the multiplayer coding harness where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud.

On AQ the agents are the real CLIs your engineers already use (Claude Code, Codex, Cursor Agent, Kimi, Grok, or plain shells), running in persistent tmux sessions on the team's VM and streamed live to the browser. Each engineer signs in with their own Claude or OpenAI account, and AQ never marks up model usage. Every workspace gets one isolated git worktree (branch ai/{id}-{slug}, automatic dependency install, one-click rebase onto main), so parallel sessions never collide. Sessions survive a closed laptop and resume from any device, including a phone. Teammates open the same workspace and watch the same live session; typing into someone else's terminal is delegated by its owner approving a control request in one click. Label a Linear issue ai-task and a workspace appears: status syncs both ways, ownership follows the assignee, and humans start agents (never auto-launched). There is no shared multi-tenant execution tier.

Where Hoplite stands out

• Lowest-friction dispatch. A Slack message, a Linear issue, or an iMessage becomes a running agent with no machine to think about; Sentry integration turns production errors into threads automatically.
• Disposable sandboxes. Parallel tasks are just more threads, with 25 to 150 pooled sandboxes on paid plans and no environment to maintain between tasks.
• Local setup imported. Bringing your agent sessions, MCP servers, and CLI configuration into the cloud at onboarding removes the classic complaint that the sandbox knows nothing about how you work.
• Verification in the loop. Agents run your tests and attach evidence to the PR, with approval gates for cautious teams.
• One bill for everything. Model usage is served and metered by the platform as credits; nobody connects a personal subscription. Pro or con depends on your procurement.

Where AQ wins

• Your infrastructure, not a vendor sandbox. AQ agents run on VMs you connect from your own cloud or on a dedicated always-on AQ-managed VM in its own isolated network, with no shared multi-tenant execution tier. See the best cloud coding agents in 2026 for the landscape.
• Your own CLIs and accounts. Hoplite's agents are Hoplite's, on models it serves and meters. AQ runs the CLIs your team already subscribes to, on their own accounts, with no markup; your configuration and muscle memory keep working.
• A session the whole team is in. Hoplite streams actions to whoever dispatched the thread. On AQ the session itself is the shared surface: teammates open the workspace, watch the same terminal, and the owner delegates input in one click. Visibility is owner-managed: team-visible, or private and shared with specific people.
• Durable workspaces. The same worktree and running session persist until the work is merged, resumable from any device, so long-running or exploratory work is not re-dispatched task by task. See always-on coding agents.
• Previews for people outside the session. Each workspace runs a live dev-server preview with viewing links that need no account; pinned comments (screenshots supported) go to the agent as prompts, and outside reviewers comment via a review link after a quick email sign-in.
• Humans stay the starters. Linear intake creates workspaces, but agents never auto-launch. For teams nervous about agents acting on production signals unprompted (Hoplite's Sentry trigger is exactly that), the default matters.

The deeper difference: a task you dispatch vs a place you work

Hoplite is built around delegation: describe the task well enough and the agent comes back with a reviewable PR. That shape is efficient for well-specified grunt work. Its weakness is everything that is not a clean task: work that needs mid-course correction, a second engineer's judgment halfway through, or an hour of back-and-forth in the same context.

It also concentrates the work on one person. A July 2026 LeadDev analysis of 25,264 agent-generated pull requests across 2,361 popular GitHub repositories found that in 79 percent of agentic PRs the same developer both reviewed and modified the agent's contribution, and only about one in eight workflows involved multiple humans. A dispatch-and-review pipeline leaves that pattern intact: the thread belongs to the dispatcher, and everyone else meets the work as a finished diff. AQ makes the opposite bet: the most useful thing a team can share is the live session itself, so a teammate can catch the wrong turn before it becomes a 40-file PR, and take the keyboard when showing beats prompting.

The two shapes can coexist. If you are choosing one place for agents to live, decide which failure mode costs more: tasks that needed steering and did not get it, or sessions that needed dispatching convenience and did not have it.

Pricing compared

Hoplite, as of September 2026: a free tier with $100 of included model usage, Pro at $99 per seat per month with 25 pooled sandboxes, Scale at $299 per seat per month with 150 pooled sandboxes, and custom Enterprise pricing. Model usage is metered as credits through the platform.

AQ has two plans. Free is a personal sandbox for one person: AQ creates a private machine in an isolated network, nothing to install, no time limit. Team is $50 per user per month early access (standard $200, billed monthly), covering VMs you connect from your own cloud or a dedicated always-on AQ-managed VM in its own isolated network, with your rate locked for your first 12 months. Inviting a teammate costs nothing until the person accepts and joins. Model usage never appears on an AQ bill: agents run on your own Claude and OpenAI accounts, and AQ never marks up model usage.

Bottom line

Hoplite is one of the better-executed dispatch platforms of 2026: sandboxes that inherit your local setup, dispatch from the tools you already use, verification evidence on the PR, and approval gates for the cautious. If your goal is to fire well-specified tasks into the cloud and review what comes back, it deserves a serious look. AQ answers a different question: where does the team work with agents, together, on machines you control, with the subscriptions you already pay for. If agent work should be a live session anyone can open, watch, and steer rather than a queue of returned PRs, that is AQ's side of the line.