AI coding agents & orchestration FAQ

Connect a GitHub account. GitHub is the substrate the platform builds on — every feature, branch, PR, and code analysis pass routes through it. Once GitHub is connected, you can browse your repos, enable code analysis on the ones you want agents to work in, and start dropping feature requests into chat.

Next is the terminal integration: install the AI Expedite Terminal app on the computer where Claude Code or Codex is already authenticated, and connect it to your workspace. That gives the orchestrator a worker it can dispatch coding tasks to.

For Ship workflows, yes. Agents that run Claude Code, Codex, tests, or builds need a real machine with a real shell, and the terminal app is how the orchestrator reaches that machine. For Discover and Launch workflows, no — those run entirely in the cloud against your connected integrations.

The terminal app is small, code-signed for macOS and Windows, and restricted to an allowlist of commands by default. See the Help Center for installation walkthroughs per OS.

A single developer can be productive in under thirty minutes: connect GitHub, install the terminal app, drop the first feature request into chat, watch the agent run. A team of three to five typically takes a few hours: invite members, set workspace permissions, connect the integrations the team relies on (Linear / Jira / Confluence / Discord), and decide which repos run code analysis.

Larger teams take longer mostly because of integration scoping — deciding which OAuth scopes to grant per integration, which repos to enable analysis on, and which workflows to standardize.

Yes. The Free plan is the standing trial: 50 credits per month and limited access across Discover, Ship, and Launch with no credit card required. Most teams use the free plan to validate the basic flow — drop a feature into chat, watch a small task complete end-to-end — before upgrading.

You can run the platform against our metered model access — there's no requirement to bring your own subscription. That said, teams that ship more than a few features a week usually find it cheaper to attach a Claude Pro / Max subscription or a ChatGPT Team seat through the terminal app. The orchestrator routes to whichever option is cheapest for each job.

Three places: the Help Center has step-by-step walkthroughs for installation and setup issues; the FAQ (this page) covers the most common product questions; the Discord community is where engineers ask questions in real time and the team responds. For procurement or security review questions, email ai@aiexpedite.com.

Yes. Every workspace surface ships with an in-app tour you can launch from the help menu. The tour walks through the kanban, the chat, the terminal integration, and the most-used workflows. You can replay any tour any time.

Yes — through the Jira and Linear integrations. Both are bidirectional, so importing the backlog also means future changes inside AI Expedite flow back into your existing system of record. If you don't use Jira or Linear, the platform's own roadmap can be the system of record.

It's the architecture you get once a feature passes through more than one specialist agent on its way from idea to shipped. A planning agent breaks the feature into a design; an implementation agent (Claude Code or Codex) writes the code; a reviewer agent checks the diff; a UI test agent verifies the change in a browser; a changelog agent drafts the launch copy.

AI Expedite is the orchestrator that keeps those agents coordinated — same backlog, same workspace, no two agents stepping on each other. See the Multi-agent delivery page for the full overview.

The orchestrator routes based on the kind of work and the agents available. Exploratory feature work goes to Claude Code; batch refactors and high-throughput jobs go to Codex; review goes to a reviewer agent tuned for diff reading; UI verification goes to the browser test agent; launch copy goes to the changelog and social drafts.

Routing is automatic by default. Teams can override it per-workflow or per-repo if they want certain work to always run on a specific agent.

Yes. Every machine connected through the terminal app becomes a worker the orchestrator can dispatch to. Five engineers each running an agent in parallel can ship five features at once, with no one developer babysitting the queue. The orchestrator distributes work across whatever workers are online and respects branch-level locking so two agents never edit the same branch at the same time.

The session surfaces in the workspace with its current state and full transcript. You can step in and chat directly with the running session, unblock it with a hint, hand the task back to the queue, or cancel it outright. Failed runs include the build / test output so the next attempt has the failure context attached.

The orchestrator does not silently retry failed runs — every retry is either initiated by you or explicitly configured in the workflow.

Yes. The agent builder lets you compose new agents from existing primitives — model selection, tool access, MCP server connections, prompt scaffolds, and approval gates. Custom agents publish into the same workspace and use the same dispatch infrastructure as the built-in ones.

CI is great for executing a known job. Multi-agent orchestration is for the case where the job needs feature-level reasoning, a live workspace, and a human in the loop on the right decisions — which CI doesn't model well. AI Expedite gives you that loop without forcing every change through a pipeline, and the workflows still produce PRs that your existing CI can run on as the final gate.

Only through the terminal app, only on machines you've explicitly connected, and only for commands the terminal has been pre-approved to run. Anything outside the allowlist surfaces a native OS-level approval dialog with the exact command before execution.

Agents do not have a background scanner that reads your filesystem; files are read into context only when the agent explicitly requests them as part of an active task.

Code analysis builds a structured index of the repo — file structure, exports, call graph, type information where available, conventions in use. The index lets agents reason about code in context: when Claude Code edits a function, the orchestrator knows about every call site; when the reviewer reads a diff, it knows what tests cover the changed lines.

Code analysis is optional. Ship works better with it on (most agents reference the index), and Launch features that need codebase context require it.

A small application that is installed on your Windows, Mac, or Linux computer. It is primarily used to interact with Claude Code or Codex and run tests against your codebase. The terminal integration lets our central agents run CLI commands on your integrated computers, so they can delegate coding tasks to your configured subscription and verify changes locally.

The terminal app is code-signed (macOS notarized, Windows Authenticode), restricted to an allowlist of commands by default, and prompts you for OS-level approval on anything outside the allowlist.

Yes. AI Expedite handles the vast majority of our day-to-day feature work, and we drop into the IDE for the occasional tweak or one-off — typically once every ten or so features.

Every feature is implemented on its own branch, so you can edit, review, or commit alongside the agent without stepping on its work. Changes flow through GitHub, so anything AI Expedite pushes shows up in your IDE on the next pull, and anything you commit locally is picked up the next time the agent runs.

If you are coding directly on a computer that is connected to AI Expedite via the terminal integration, click 'Disconnect from cloud' in the terminal app first. Otherwise, background sync may stash your in-progress changes and switch branches while you are working.

Most features go from idea to merged code in three steps:

1. Pick a workflow. From the chat bar, choose Auto, Requirements, Design, or Quick Fix based on how much structure you want around the work.

2. Ask chat to create a feature request. AI Expedite turns your prompt into a request and drops it into Roadmap → Kanban so you can track it alongside everything else in flight.

3. Approve each step as it runs. Requirements, designs, implementation plans, and code changes are all linked — the output of one step becomes the input for the next, so context never gets lost between handoffs. The agent uses Claude Code or Codex (whichever the workflow routes to), runs the resulting tests, and opens a PR you can review and merge.

It selects the workflow your feature will follow.

Auto: AI Expedite chooses the best workflow (Beta).

Requirements: Best for larger work that needs formal requirements.

Design: Best when you want a design step before implementation (Recommended).

Quick Fix: Best for small, straightforward changes.

More structure means more approval gates but higher-quality output for non-trivial features. Quick Fix is appropriate for things like 'rename this variable everywhere' or 'add a unit test for X'.

No, but some features require it, such as Launch features and human-readable documents. Ship also works better with code analysis enabled because our agents reference the metadata to help Claude Code and Codex follow your coding standards. The code analysis index is the substrate for AI code review accuracy and roadmap effort estimation — without it those workflows fall back to less informed defaults.

Yes. The terminal integration is designed with user approval for sensitive operations, HMAC-signed command verification, secure device registration, and automated PII redaction in session logs. Commands are normalized before allowlist matching, so a chained operator or newline cannot smuggle a second command through. The signed binary is source-available on GitHub so you can audit it before you install it.

Yes, GitHub is currently required. AI Expedite is architected to support other version control systems in the future. Request additional integrations at ai@aiexpedite.com.

Usually, no. Human-readable code documentation works great, but we only recommend enabling it when you are ramping up a technical team or actively trying to understand a codebase.

Because these documents update on every code commit, the cost can add up over time. If you do not need continuously refreshed, easy-to-read code documentation, leave it off and enable it only when the extra context is worth the ongoing cost.

Yes. Every agent-produced change lands as a GitHub PR, so your existing CI runs against it the same way it runs against human-authored PRs. The orchestrator also reads CI status — failing builds and tests come back into the agent as failure context so it can fix forward.

Discover helps turn workspace activity, customer feedback, analytics, errors, and connected tools into prioritized product and roadmap insights. It's the front half of the loop that ends with Ship and Launch — the part where signals become a roadmap.

Discover can help identify customer pain points, feature opportunities, roadmap themes, error trends, analytics patterns, and suggested next actions. Each insight carries the underlying evidence — usage funnel screenshots, support ticket excerpts, error stack traces — so the PM can verify the recommendation before accepting it.

Discover uses the integrations you connect, such as Jira, Linear, Discord, Google Analytics, Google Search Console, GCP Error Reporting, and workspace documents. Meeting transcripts (from Google Calendar / Zoom recordings) are an additional source if you grant access. The system clusters across sources so a theme that appears in three different places is weighted higher than one with a single mention.

No. Discover can recommend roadmap items and priorities, but you stay in control of what gets accepted, edited, or added to your roadmap. The system can re-score existing items in the background as new signals land, but adding, removing, or re-ordering items requires explicit approval.

Yes. Discover can help produce summaries, recommendations, research notes, and planning documents based on connected workspace context. Some document features work best when code analysis and relevant integrations are enabled — the more grounding the system has, the more specific the output.

The system ingests your connected signals, clusters them into themes, and proposes roadmap items with the supporting evidence linked. Each item carries a score (reach × severity × confidence × effort) that's a recommendation, not a verdict. You see the inputs and can adjust any of them; items that pass your threshold land on the kanban board after your approval. See the AI roadmap generation landing page for the full walkthrough.

Partly. Signal ingestion and theme clustering work without code analysis. What requires code analysis is effort estimation — the system needs to read your repo to estimate how big a change actually is, instead of just guessing from the user-facing description. Teams that want full prioritization scoring should enable code analysis on the relevant repos.

Yes. Each insight and recommendation can be exported as a shareable document — PDF or web link — that includes the supporting evidence. Useful for board reviews, customer interviews, or executive readouts.

Launch helps plan, create, schedule, and track go-to-market activity, with a focus on social content posting, ad workflows, and performance tracking. It closes the loop between shipped code and growth: every changelog entry can become a launch campaign, and every campaign's performance feeds back into future drafts.

Launch supports social workflows across LinkedIn, X / Twitter, Facebook, Instagram, YouTube, Threads, Reddit, and other channels as integrations become available. TikTok integration is in development.

Each channel gets its own native draft — different structure, length, hashtag conventions, and tone. The system never just cross-posts the same copy across all of them.

Yes. Launch can help draft posts, repurpose product updates into channel-specific content, and prepare campaigns for scheduling or review. The source material comes from your shipped feature work — changelog entries, feature requests, PR descriptions — so the copy references what actually shipped, not generic marketing language. See the AI social post generation landing page for the full overview.

Yes. AI Expedite can generate AI images and videos for launch activities. We also have a beta feature that can generate images sourced from your website to keep the brand visuals coherent.

Not yet, but we are working on it. We are building this for ourselves first, and then we will roll it out to customers.

Yes. Launch supports Google Ads workflows, including campaign planning, creative support, and performance tracking where your account is connected. The system tracks ROI, conversions, and campaign metrics through the Analytics integration and uses past performance data to inform future creative drafts.

Yes. Launch is designed for planning and scheduling social content. Availability depends on the connected channel and its current integration status. The default is review-before-publish: drafts wait for your approval, then publish on the schedule you set. You can opt into auto-publish per channel as you build trust with the drafts.

Yes. Launch can use connected analytics, ads, and social integrations to monitor performance and use your past results when drafting post recommendations. This gets better over time as Launch continues learning from what performs best for your audience. The performance loop is what tunes future drafts — post formats and headline structures that drove conversions get weighted higher in subsequent suggestions.

Yes. The changelog-to-launch automation publishes entries to your hosted changelog page (or a configured subdomain), GitHub releases, or any webhook endpoint, and stages the launch campaign in the same flow. See the changelog-to-launch landing page for the full overview.

Credits power AI execution across Discover, Ship, and Launch — for example, generating insights, running coding agents, creating launch content, and other model-driven work. Browsing the app, manually editing documents, or changing billing settings does not use credits. Most of the cost of a typical feature is in agent orchestration and review; the heavy model work runs on your attached subscription where possible.

By default, AI Expedite pauses new AI work when you hit your included limit. You can buy a top-up or enable Auto-Reload with a monthly cap.

Yes. Overage billing is off by default on self-serve plans. Auto-Reload can be enabled with a maximum monthly cap so spend is bounded.

Self-serve pricing is per seat, not workspace. Code analysis and scheduled activities are charged to the workspace owner (edit in Settings → Workspace → Members). That keeps cost predictable for teams while letting each engineer drive agent work from their own account.

Monthly plans can be upgraded anytime, with upgrades taking effect immediately and downgrades applying at the next renewal. Annual plans are billed up front for 12 months.

Included monthly credits reset every billing cycle and do not roll over. Prepaid top-up credits roll over for 60 days before expiring.

Annual billing prepays 12 months at a 15% discount. Your included credits still refresh each month rather than pooling for the whole year. Annual customers can still buy top-ups or enable Auto-Reload if they exceed the monthly allotment.

The Free plan is the standing trial: 50 credits per month and limited access across Discover, Ship, and Launch with no credit card required.

For refund questions, email ai@aiexpedite.com — we review requests on a case-by-case basis. Generally, renewals, top-up credits, and consumed overages are non-refundable, but reach out and we'll work with you on a fair resolution.

Yes. Top-ups are best for occasional spikes. If you need extra credits every month, upgrading is usually the better value because it reduces per-credit cost.

Self-serve plans are card-based. If you need annual invoicing, procurement support, or a larger deployment, contact us at ai@aiexpedite.com.

Legacy customers keep their current plan through the end of the current billing term and receive a bridge offer before moving to the new pricing model.