The agentic coding stack
For developers who want AI across the whole workflow, from keystroke to delegated task.
Three tools, three altitudes. Completions and workflow integration where you already work, an AI-first editor for the code you actively shape, and an autonomous agent for the tasks you would rather assign than perform. Together they cover assistance, collaboration and delegation, which is the full spectrum of what coding AI now offers.
The stack, step by step
- 01
GitHub Copilot
The baseline: completions, chat and review woven through GitHub and every major editor.
Swap options- Tabnine when the organisation requires on-premises or air-gapped deployment with zero code retention See the comparison
Copilot is the baseline woven through GitHub and every major editor; Cursor is the dedicated environment for the code you actively shape, with an indexed codebase and agentic multi-file edits. Same repository, different altitude: ambient completions everywhere, then a deep session where the real changes happen.
- 02
Cursor
The editor: deep codebase context and agentic multi-file edits for the work you steer by hand.
Swap options- Devin Desktop when predictable quota-refresh pricing suits heavy agent use better than credit metering See the comparison
Cursor is collaboration you steer by hand; Claude Code is delegation, with whole tasks briefed like tickets and executed against the repository from the terminal while you review. The editor covers the work you own, the agent absorbs the work you assign, and both read the same codebase.
- 03
Claude Code
The delegate: whole tasks briefed like tickets and executed against the repository while you review.
Swap options- OpenAI Codex when the team already pays for ChatGPT and wants delegation bundled into that plan rather than metered separately See the comparison
- Devin Cloud when tasks are well-scoped backlog items to assign outright, with review capacity to absorb the output See the comparison
What it costs
Individual tiers of all three are affordable relative to any developer's time; the honest budget grows with usage-based model consumption, which tracks how much you actually delegate.
| Tool | Entry tier | What drives cost up |
|---|---|---|
| GitHub Copilot | Free tier + paid plans | Free ($0); Pro ($10/mo, includes $15/1,500 credits); Pro+ ($39/mo, 7,000 credits); Business ($19/user/mo); Enterprise ($39/user/mo). 1 AI Credit = $0.01. |
| Cursor | Free tier + paid plans | Hobby (free); Pro ($20/mo); Pro+ ($60/mo, 3× usage); Ultra ($200/mo, 20× usage); Teams ($40/user/mo). |
| Claude Code | Paid only | Included with Claude Pro ($20/mo), Max ($100/$200/mo) or Team Premium; or pay per token via the API (Opus 4.8 $5/$25 per million input/output tokens). |
Compare the members
Written comparisons between these tools and their nearest substitutes.
Built for
The three tiers of this stack
Ready
The starter coding stack
For developers getting their first leverage from AI, without changing editors or spending anything.
Competitive · this stack
The agentic coding stack
For developers who want AI across the whole workflow, from keystroke to delegated task.
World-Class
The delegation engineering stack
For engineering teams treating AI as headcount: work assigned, executed and reviewed, not just assisted.
Common questions
What does this stack actually cost per month?
Two of the three tools have a free tier to start; Claude Code is paid from day one. The 03 COSTS table above breaks down each tool's plans. Three meters climb with use: Copilot's AI credits, Cursor's usage-based credit metering, and Claude Code's per-token billing. The token and credit meters bite hardest as you shift from steering code to delegating it, which is itself a sign the delegation is paying off. Budget for consumption, not just the base plans.
Do I need all three tools from day one?
No. Copilot alone is the low-friction baseline, and Cursor alone spans assisted work and much of what you would delegate; either carries an early developer. The numbered steps double as the adoption order: start with Copilot, since it needs no migration; add Cursor when agentic multi-file edits justify changing editors; bring in Claude Code once you genuinely brief work rather than steer it.
I already use GitHub Copilot. What changes?
Your Copilot subscription already covers the baseline: completions, chat and review woven through GitHub and whatever editor you use. That stays, and you can defer the delegation altitude until you actually brief work. What this stack adds is what Copilot does not reach: Cursor's indexed-codebase context and agentic multi-file edits for code you steer by hand, and Claude Code executing whole tasks against the repository while you review.
Where do these tools overlap, and which wins?
Copilot and Cursor both put AI in your editor, the one real overlap here. The dividing rule is depth. Copilot wins as the ambient layer: completions and chat woven through GitHub and whatever editor you already open, no migration required. Cursor wins for the deep session, where an indexed codebase and agentic multi-file edits do the real reshaping. The GitHub Copilot and Cursor comparison, linked above, covers the finer calls.
When do I outgrow this stack?
Two signals. When delegation stops being occasional and becomes how most of the work ships, so you need several agents running in parallel and a way to orchestrate and review them rather than one task at a time. And when a whole team, not a lone developer, has to govern that autonomous output with shared controls and audit. Both point to the world-class tier, the delegation engineering stack.
What can I safely put into these tools?
The strictest member sets the floor for shared work, and here it stays low until you configure each tool. Keep proprietary code out until each excludes training: Copilot's Business or Enterprise tier, Cursor's team privacy mode, and Claude Code through commercial or API access rather than a consumer plan. Set budget caps on the agents, and review everything they generate for correctness and security before it ships.
Before sharing confidential or personal data, check this tool's data-governance and training policies. They differ between providers and can change.
Last checked: July 2026
Where to start
Not sure what to adopt first?
Five quick questions about your job, task and constraints. We'll suggest your top three tools, plus the one to try first.
Tool facts last checked July 2026