How Macroscope's AI Coding Agent Works: Ask Questions, Write Code, Ship PRs
Macroscope Agent is an AI coding agent that answers questions about your codebase, writes code, opens pull requests, and connects to Slack, Jira, Linear, PostHog, Sentry, and any MCP server.
Macroscope Agent is an AI coding agent that answers questions about your codebase, writes code, opens pull requests, and triggers on schedules or events. It is the part of Macroscope that goes beyond AI code review — when the review surfaces a problem, the AI coding agent can investigate the root cause across services, draft the fix, and ship the PR. Reach it from Slack, GitHub PR threads, the Macroscope web app, scheduled Macros, or a webhook API.
TL;DR — What is Macroscope Agent?
- What: An AI coding agent built into Macroscope. Answers questions about your codebase, writes code, opens PRs, and runs scheduled or event-triggered automations.
- Where you use it: Slack DMs and channel mentions, GitHub PR comments, the Macroscope web app, Macros (scheduled or event-triggered), and a webhook API.
- What it can read: Your codebase, git history, GitHub API, plus connected services — Jira, Linear, BigQuery, PostHog, Amplitude, LaunchDarkly, Sentry, GCP Cloud Logging, and any MCP server.
- What it can do: Create branches, write code, open PRs, update branches, add reviewers, create issues in Jira/Linear, generate diagrams.
- Pricing: Usage-based. Agent is $0.01/credit with 1,000 free credits/month per workspace (~$10 value). Code Review is $0.05/KB; Status is $0.05/commit.
- Free usage: Every new workspace gets $100 in free usage to start.
What Is Macroscope Agent?
Macroscope Agent is an AI coding agent that lives inside your Macroscope workspace. It started as the chat surface for asking questions about your codebase — Ask Macroscope Anything. As more capabilities were added (code generation, branch creation, PR shipping, integrations with the rest of your SDLC), it stopped being a chat tool and became a full AI coding agent.
So this is not a thin chat widget over an LLM. The Macroscope agent has access to your codebase via the same AST-based reference graph that powers Macroscope's AI code review. It runs read-only queries against GitHub, opens PRs, talks to Jira and Linear, queries production analytics in PostHog and Amplitude, reads errors from Sentry, and can be extended with any MCP server you connect to it.
If you have asked an AI coding agent a question about your codebase before and gotten a confidently wrong answer because it could only see one file at a time — that is the gap Macroscope Agent closes. Real codebase context, real production context, real action.
What Macroscope Agent Can Do
Macroscope Agent operates on three axes: what it can read, what it can write, and where it can run. Each axis is what separates a useful AI coding agent from an autocomplete demo.
What it can read
Out of the box, every Macroscope agent can:
- Search your codebase using the AST-based reference graph
- Analyze git history (who changed what, when, and why)
- Look up pull requests and review activity
- Run read-only queries against the GitHub API (PR metadata, issue history, workflow runs, repo stats)
- Pull contributor metrics
- Retrieve sprint reports
- Search Macroscope's own docs
- Search the public web when context is missing
What it can write
The Macroscope agent is not a read-only AI code assistant. It can:
- Create branches in your repositories
- Write code (full diffs, multi-file changes, new files)
- Open pull requests with descriptions and linked context
- Update existing branches
- Add reviewers
- Create issues in Jira or Linear
- Generate images (architecture diagrams, sequence diagrams)
Connected services
When you connect external services in Macroscope, the AI coding agent can use them as tools:
- Jira — look up issues, track project progress, create issues
- Linear — look up issues, track project progress, create issues
- BigQuery — browse datasets, inspect schemas, run read-only SQL with methodology summaries
- PostHog — query analytics events, inspect feature flags, run HogQL, search session recordings
- Amplitude (beta) — event segmentation, funnels, retention, user event streams
- LaunchDarkly — list projects, environments, feature flags, and targeting rules
- Sentry (beta) — search issues and investigate crashes and exceptions
- GCP Cloud Logging — search and filter logs across connected projects
- MCP Connectors (beta) — add custom MCP servers like Datadog, PagerDuty, or your internal APIs
This is the difference between an AI code assistant and an AI coding agent. An assistant answers from the file you are looking at. An agent reaches into the rest of your stack.
How Does Macroscope Agent Access Your Codebase?
When the Macroscope GitHub App is installed on a repository, Macroscope builds an Abstract Syntax Tree (AST) of every supported file using language-specific codewalkers — Go, TypeScript, JavaScript, Python, Java, Kotlin, Swift, Rust, Ruby, Elixir, and Starlark. These codewalkers construct a reference graph: every function, class, type, and variable, and every place each one is defined and used.
When you ask the AI coding agent a question, it does not grep blindly. It traverses the reference graph. Ask "where do we validate webhook signatures?" and Macroscope walks every call site of your verification function and the hierarchy around it. Ask "what changed in the payments service this week?" and it filters git history through the same graph.
This is why the Macroscope agent can answer cross-file questions that single-file AI code assistants cannot — it has structural understanding, not just text proximity.
Where Can You Use Macroscope Agent?
The Macroscope AI coding agent is intentionally surfaced in places engineers already work, not behind a separate chat tab nobody opens.
In Slack
Direct-message @Macroscope or @-mention it in any channel. The agent reads thread context, attached images, screenshots, and PDFs. Ask it to investigate a Sentry alert that just landed in your alerts channel and it will read the alert, query Sentry, look at the relevant code, and reply in-thread with a root-cause analysis or a draft fix PR.
Group DMs are supported, so you can pull the AI coding agent into a triage thread with other engineers and treat it as another participant.
In GitHub PR threads
@-mention @macroscope on any pull request, or reply to one of its inline review comments. The agent reads the PR diff, the surrounding code, and any linked tickets — then answers, suggests, or fixes. Replying "fix it for me" to a review comment hands the issue to the agent's Fix It For Me workflow: branch, fix, PR, CI loop, optional auto-merge.
In the Macroscope web app
The Macroscope web app exposes the AI coding agent directly so you can run multi-step queries, open PRs from chat, and inspect the agent's tool calls and reasoning.
In Macros (scheduled or event-triggered automation)
Macros let the AI coding agent run on a schedule (daily, weekly, biweekly) or on an event (new commit, new pull request) — without anyone asking. More on this below.
Via the webhook API
Send a POST to https://macrohook.macroscope.com/api/v1/workspaces/{workspaceType}/{workspaceId}/query-agent-webhook-trigger with a query and a response destination, and the AI coding agent runs asynchronously. The result is delivered to a Slack channel, a Slack thread, or your own webhook URL.
Macros: Scheduled and Event-Driven AI Coding Agent Automation
Macros are the AI coding agent running on autopilot. You write the prompt in plain English, pick a trigger, pick a delivery target, and it runs.
Triggers
- Scheduled — daily, weekly, or biweekly
- New commit — fires whenever a new commit lands in a matching branch
- New pull request — fires whenever a new PR is opened
Each event-triggered macro receives the relevant metadata (commit info, PR info) as context.
Filters
- Repo filter — only run for repos matching a glob pattern
- Branch filter — restrict commit-based macros to specific branches
- Author filter — restrict to commits or PRs from specific authors
Delivery targets
- A Slack channel
- A specific Slack thread (so the agent's response lands in context, not as a new message)
- A webhook URL (validated against a workspace allowlist at configuration and execution time)
Example macros teams run
- Daily production health sweep — every morning, the AI coding agent queries GCP Cloud Logging, clusters the most frequent errors over the last 24 hours, identifies which services are affected, and drafts PRs for the top issues.
- PR ticket-link enforcement — every new PR, the AI coding agent checks for a linked Jira or Linear ticket, posts a comment if missing, and links the matching ticket if it can find one.
- Weekly changelog draft — every Friday, the AI coding agent walks the week's merged PRs and drafts a customer-facing changelog post, posted to Slack for the team to edit.
- Sentry auto-fix — whenever a Sentry alert lands in
#alerts, the AI coding agent investigates the error, walks the codebase, and pushes a draft fix PR. - Customer support draft — every new Pylon ticket fires a webhook to Macroscope. The AI coding agent reads the customer's question, decides if it is a bug, a product question, or a feature request, and drafts the appropriate response or PR.
Anyone with Macroscope web-app access can create a macro. There is no separate Macros pricing — the same usage meters apply.
Macroscope Agent vs GitHub Copilot
GitHub Copilot is an AI code assistant. Macroscope Agent is an AI coding agent. The difference matters when you actually try to ship work with them.
| Capability | Macroscope Agent | GitHub Copilot |
|---|---|---|
| Surface | Slack, GitHub PRs, web app, Macros, webhook API | IDE editor, GitHub PR (Copilot Workspace) |
| Codebase access | AST reference graph across the full repo | Active file + nearby context |
| Cross-file reasoning | Yes (reference graph) | Limited |
| Open PRs autonomously | Yes | Yes (Copilot Workspace) |
| Validate fixes against CI | Yes (Fix It For Me iterates until tests pass) | No automatic CI loop |
| Connected tools | Jira, Linear, BigQuery, PostHog, Amplitude, LaunchDarkly, Sentry, GCP Logging, MCP servers | None native |
| Scheduled / event-driven | Yes (Macros) | No |
| Webhook API | Yes | No public agent webhook |
| Production observability access | Yes (Sentry, GCP Logging, PostHog, Amplitude) | No |
| Pricing model | Usage-based, included with Macroscope | $10–$39/seat/month |
Copilot is excellent at editor-time autocomplete and inline suggestions. Macroscope Agent is built for the work that happens around the editor: investigating production issues, answering questions in Slack, drafting PRs from a ticket, running scheduled checks across the codebase. Different surfaces, different jobs.
Macroscope Agent vs Cursor
Cursor is an AI-powered IDE. Macroscope Agent is an AI coding agent that lives in your collaboration surfaces (Slack, GitHub, web app, automations). They are not direct competitors and many teams use both.
| Capability | Macroscope Agent | Cursor |
|---|---|---|
| Where it runs | Slack, GitHub PRs, web app, Macros, webhook API | IDE (forked VS Code) |
| Codebase context | Full AST reference graph | Embeddings + agentic search of repo |
| Where work lands | A real PR in GitHub | Edits in your editor |
| Connected tools | Jira, Linear, BigQuery, PostHog, Amplitude, LaunchDarkly, Sentry, GCP Logging, MCP servers | MCP servers (in IDE) |
| Scheduled automation | Yes (Macros) | No |
| Event-triggered automation | Yes (commit / PR triggers) | No |
| Used asynchronously by your team | Yes — anyone in Slack | No — single-developer IDE |
Cursor is for the developer typing in the editor. Macroscope Agent is for the team operating asynchronously across Slack, GitHub, and production. The agent does not need an engineer to be at their desk to work — it can run a daily sweep and have draft PRs waiting.
Macroscope Agent vs CodeRabbit
CodeRabbit is an AI code review tool focused on PR comments and learnings. Macroscope Agent is broader: it includes AI code review, but it also writes code, opens PRs, runs in Slack, and runs on schedules.
| Capability | Macroscope Agent | CodeRabbit |
|---|---|---|
| AI code review | Yes (Macroscope Code Review v3, 98% precision) | Yes |
| AI coding agent (writes code, opens PRs) | Yes | Limited (1-click commits, AI suggestions in review threads) |
| Slack as a primary surface | Yes (DMs, channels, threads) | Limited |
| Scheduled automations (Macros) | Yes | No |
| Event-triggered automations | Yes (commit / PR) | No |
| Webhook API for triggering the agent | Yes | No |
| Connected tools (Sentry, PostHog, BigQuery, etc.) | Yes | Limited |
| MCP server support | Yes | Limited |
For teams already shopping for CodeRabbit alternatives, Macroscope is the broader product: AI code review plus an AI coding agent that takes action across the rest of your stack. Compare in detail at Macroscope vs CodeRabbit.
Macroscope Agent vs Greptile Chat
Greptile Chat is a codebase Q&A tool — ask questions, get answers grounded in your repo. It is a chat surface, not an AI coding agent. It does not write code, does not open PRs, does not run on schedules.
| Capability | Macroscope Agent | Greptile Chat |
|---|---|---|
| Codebase Q&A | Yes | Yes |
| Writes code | Yes | No |
| Opens PRs | Yes | No |
| Slack integration | Deep (DMs, channels, threads, attachments) | Limited (Chat only) |
| Scheduled automation | Yes (Macros) | No |
| Connected tools beyond GitHub | Jira, Linear, BigQuery, PostHog, Amplitude, LaunchDarkly, Sentry, GCP Logging, MCP | Limited |
| Pricing | Included with Macroscope (no per-user fee) | $20/user/month on top of $30/seat code review |
Teams searching for Greptile alternatives often want what Greptile Chat does not do — write code, open PRs, run automations. The Macroscope AI coding agent is built for that. See the full Macroscope vs Greptile comparison.
How Macroscope Agent Fits the GitHub Code Review Workflow
A standard GitHub PR review with the Macroscope AI coding agent in the loop:
- Developer opens a PR on GitHub.
- Macroscope's AI code review runs automatically — Correctness check, custom Check Run Agents, PR summary.
- The reviewer (or the agent) finds something to discuss. A teammate can @-mention
@macroscopeon the PR with a question like "is this consistent with how we handle webhook retries elsewhere?" The AI coding agent walks the codebase and answers in-thread. - A bug is found. Reply "fix it for me" — the agent creates a branch, writes the fix, opens a PR, runs CI, iterates on failures until tests pass.
- Linked Jira or Linear ticket? The agent pulls the ticket description and acceptance criteria into the review and the fix.
This makes the Macroscope AI coding agent the connective tissue across GitHub code review, GitHub PR review, ticket trackers, and production observability — not a separate place to remember to go.
How Customers Use the Macroscope AI Coding Agent
A few real patterns we have seen:
- Triage Sentry alerts in Slack. A Slack channel collects Sentry alerts. A Macro picks each one up, the AI coding agent investigates the stack trace against the codebase, and posts a root-cause analysis in-thread. For straightforward issues, the agent attaches a draft fix PR.
- Auto-link tickets to PRs. A new-PR Macro asks the AI coding agent: "does this PR have a Jira ticket linked? If not, search Jira for the most likely ticket and post a comment with the link." Cuts hand-linking work to zero.
- Daily codebase digest. A scheduled Macro asks the AI coding agent to summarize the last 24 hours of merged PRs by area of the codebase and post it to
#engineering. The team reads one Slack message instead of scrolling every PR title. - Customer support draft. A webhook from Pylon hands new tickets to the AI coding agent. It classifies the ticket (bug, product question, feature request), writes the draft response, and either opens a fix PR or files a Linear issue.
- PR risk summary. A Macro on every PR runs the AI coding agent against the diff: "what production data does this touch? Which tests cover it? What is the blast radius?" The summary lands as a PR comment so reviewers know what to focus on.
What Macroscope Agent Costs
Macroscope uses usage-based pricing. There is no per-seat, per-user, or per-integration fee.
The three published meters:
- Code Review: $0.05 per KB reviewed (10 KB minimum per review = $0.50 floor)
- Status (commit processing): $0.05 per commit
- Agent: $0.01 per credit, with 1,000 free Agent credits per month per workspace (~$10 of free Agent usage every month)
If an agent action triggers commit processing or a code review, the standard Code Review or Status meter applies on top. There are no separate per-Macro, per-API-call, or per-integration fees beyond Agent credits.
Default spend controls: $10 per review cap, $50 per pull request cap. Both adjustable. Every new workspace gets $100 in free usage to start, which does not expire.
This matters for AI coding agent workloads specifically. As more code is written by agents (Copilot, Cursor, Claude Code), per-seat AI tools stay flat in cost but their per-developer review caps blow out. Macroscope's usage-based meter scales with the work, not with the seat count.
How to Get Started With Macroscope Agent
- Install Macroscope at macroscope.com. Connect the GitHub App to your repos.
- Configure Status (Product Overview, Sprint Cadence, backfill). Status configuration is the prerequisite that lets the AI coding agent reason about your team and your work.
- Connect integrations at Settings → Connections — Jira, Linear, BigQuery, PostHog, Amplitude, LaunchDarkly, Sentry, GCP Cloud Logging, or MCP servers.
- Try the agent. DM @Macroscope in Slack with a question like "where do we handle webhook retries?" or @-mention it on a PR.
- Set up your first Macro. A daily codebase digest is a good starter — it makes the AI coding agent's value visible to the whole team without anyone having to remember to use it.
Book a demo if you want a guided walk-through with your real codebase.
Frequently Asked Questions
What is Macroscope Agent?
Macroscope Agent is an AI coding agent built into Macroscope. It can answer questions about your codebase, write code, open pull requests, query connected services (Jira, Linear, BigQuery, PostHog, Amplitude, LaunchDarkly, Sentry, GCP Cloud Logging, MCP servers), and run on a schedule or in response to events. You reach it from Slack, GitHub PR threads, the Macroscope web app, scheduled Macros, or a webhook API.
How is Macroscope Agent different from GitHub Copilot?
Copilot is an AI code assistant that runs in your editor and helps you type. Macroscope Agent is an AI coding agent that runs in Slack, GitHub PRs, scheduled Macros, and a webhook API — it can investigate questions across the codebase, draft and open PRs, validate fixes against your CI, and connect to production observability tools like Sentry and PostHog. Copilot is for the developer editing code. Macroscope Agent is for the team operating asynchronously around the code.
How is Macroscope Agent different from Cursor?
Cursor is an AI-powered IDE. Macroscope Agent runs outside the IDE — in Slack, GitHub, the web app, scheduled Macros, and a webhook API. Cursor is for one developer at a desk. Macroscope Agent is for the whole team operating asynchronously across Slack and GitHub, including production observability access (Sentry, PostHog, GCP Logging) that an IDE-based agent does not have. Many teams use both.
Is Macroscope Agent a CodeRabbit alternative?
For teams evaluating CodeRabbit alternatives, Macroscope is the broader product. CodeRabbit focuses on AI code review with 1-click commits and AI suggestions inside review threads. Macroscope includes AI code review (Macroscope Code Review v3, 98% precision) plus a full AI coding agent that writes code, opens PRs, runs in Slack, runs on schedules, and connects to your production tools. See the full Macroscope vs CodeRabbit comparison.
Is Macroscope Agent a Greptile alternative?
For teams evaluating Greptile alternatives, the difference is action vs. answers. Greptile Chat answers questions about your codebase. Macroscope Agent answers questions, writes code, opens PRs, runs in Slack, runs on schedules, and connects to Jira, Linear, Sentry, PostHog, and MCP servers. See the full Macroscope vs Greptile comparison.
How does Macroscope Agent access my codebase?
Macroscope's codewalkers parse every supported file in your repository into an Abstract Syntax Tree and build a reference graph across functions, classes, and types. The AI coding agent uses this graph for cross-file reasoning — it does not just grep, it walks the structural relationships. Supported languages: Go, TypeScript, JavaScript, Python, Java, Kotlin, Swift, Rust, Ruby, Elixir, and Starlark.
Can Macroscope Agent open pull requests?
Yes. The AI coding agent can create branches, write code, open pull requests, update branches, and add reviewers. When it fixes a bug found during AI code review, it uses the Fix It For Me workflow — open a PR, run your CI, iterate on failures until tests pass.
What integrations does Macroscope Agent support?
Native connected services: Jira, Linear, BigQuery, PostHog, Amplitude (beta), LaunchDarkly, Sentry (beta), and GCP Cloud Logging. MCP Connectors (beta) let you add any MCP server — Datadog, PagerDuty, your internal APIs, anything that speaks MCP. Authentication options include header-based and OAuth, with per-tool access controls.
What are Macros and how do they relate to the AI coding agent?
Macros are scheduled or event-triggered runs of the Macroscope AI coding agent. Triggers: daily / weekly / biweekly schedules, new commit, or new pull request. Filters: repo glob, branch, author. Delivery: Slack channel, Slack thread, or webhook. Use them for daily production health sweeps, PR ticket-link enforcement, weekly changelog drafts, Sentry auto-triage, and customer support drafting.
Does Macroscope Agent have an API?
Yes. POST to https://macrohook.macroscope.com/api/v1/workspaces/{workspaceType}/{workspaceId}/query-agent-webhook-trigger with a query and a response destination (Slack channel, Slack thread, or webhook URL). The AI coding agent runs asynchronously and delivers the response when it is ready. Auth is via the X-Webhook-Secret header.
Can Macroscope Agent fix bugs found in GitHub PR review?
Yes. When Macroscope's AI code review flags a bug in a GitHub pull request, replying "fix it for me" hands the issue to the AI coding agent. It creates a fix branch, writes the fix, opens a PR, runs your GitHub Actions, reads the failure logs if CI fails, and iterates until tests pass. Optional auto-merge on green. This is the Fix It For Me workflow described at Fix It For Me.
How much does Macroscope Agent cost?
Macroscope's three published usage meters are $0.05 per KB reviewed (Code Review, $0.50 floor per review), $0.05 per processed commit (Status), and $0.01 per Agent credit. Every workspace gets 1,000 free Agent credits per month (~$10 of free Agent usage). There are no per-seat fees. Default spend caps are $10 per review and $50 per pull request, both adjustable. New workspaces get $100 in free usage to start.
Does Macroscope Agent work with GitLab?
No. Macroscope currently supports GitHub only, including GitHub Enterprise. GitLab and Bitbucket support have not been announced. For teams on GitLab evaluating an AI coding agent, this is the gating constraint.
Is Macroscope Agent secure to use on private code?
Macroscope is SOC 2 compliant. The AI coding agent uses scoped GitHub App permissions, runs read-only API calls against external integrations by default, and validates webhook URLs against a workspace allowlist at both configuration and execution time. Customer code is not used for model training.