Cursor Projects: When Freelancers Should Stop Living in One-Off Chats

Freelancers who code (or ship client MVPs) drown in disposable Cursor chats. Features, migrations, and “gardening” work need weeks of context. Cursor’s Projects (10 Sep 2026 beta) adds a coordinator + subagents + shared context. You need a buy/skip by job size — not another IDE war.

This is a project-vs-chat scope sheet for people who bill for shipped code. Cursor’s own productivity stats stay attributed to Cursor; we are not turning them into freelance earnings claims.

Affiliate disclosure: some Cursor or tool links may be affiliate or referral later. Plan prices below are from cursor.com/pricing as of 21 Sep 2026 (SAST) — re-check checkout. Projects is in beta; UI and plan gates can move. Not a guarantee of faster merges, clients, or income.

What Projects is (coordinator directs; you still review code)

Projects is Cursor’s layer for larger bodies of work — a feature, a migration, or a full app. You chat with a coordinator agent. The coordinator does not write all the code itself; it directs other agents that do. Cursor’s framing: because it delegates, it stays responsive while subagents run.

Three building blocks from Cursor’s launch post:

1. Cloud by default, local when needed — a Project can keep working when you close the laptop; local agents spin up when something must run on your machine.
2. Shared context — Project files sync across the agents on that workstream: research, artifacts, how you like tests run. You stop re-onboarding a fresh chat every Monday.
3. Subscriptions — the coordinator can watch Slack, run on a schedule, or follow PRs (including CI fixes) without waiting for a fresh prompt.

Projects is rolling out in beta. Start from the left-hand nav, describe the body of work, and let the coordinator take the first planning pass — then you steer.

This is orchestration, not “which model is smartest.” Model roundups stay separate: ChatGPT vs Claude vs Gemini (2026).

Three patterns — feature / migration / gardening — mapped to freelance SOWs

Cursor describes three patterns their own engineers use. Map them to freelance statements of work:

Feature work — multi-PR feature with research → plan → parallel implement/test → local try → monitor after ship. Freelance SOW fit: “Build X with milestones A/B/C across several PRs.” Put acceptance tests in the SOW so agent runs have a finish line.

Migrations — easy to start, hard to finish (framework swaps, styling system changes across many files). Freelance SOW fit: “Migrate module set M with incremental PRs; you review early PRs closely, then sample.” Price for review time, not fantasy “agents finish alone.”

Gardening — never-ending quality work: lint debt, design-system drift, CI flakes. Freelance SOW fit: retainer or weekly hours cap for “keep main healthy,” with explicit merge rules. Do not sell unlimited gardening as a fixed fee unless the cap is brutal.

Cursor reports internal productivity lifts for Project-heavy users (including a “merge more PRs” claim on their blog). Attribute that to Cursor’s own usage — do not paste it into your sales page as guaranteed freelance throughput.

If you are still learning how to sell AI-assisted build work at all, start here first: AI freelance for beginners.

When a normal chat is still better

Stay in a normal Cursor chat when:

• The fix is one file or a tight CSS tweak
• You need a 10-minute explanation, not a multi-day workstream
• The repo is unfamiliar and you are still mapping it by hand
• The client has not approved AI-assisted implementation in writing
• You are pair-debugging a production fire and need tight human control

Projects shine when context must survive longer than one evening. Chats shine when the blast radius is small. Using Projects for a one-line typo is ceremony. Using a naked chat for a six-week migration is amnesia.

Decks and proposals are a different deliverable lane — complementary, not competing: Gamma AI for client decks.

Scope and money — write the SOW so agent runs don’t eat the fee

Do not invent token-dollar tables. Do build habits:

1. Name the workstream in the SOW (“Auth refactor Project”) so scope is a noun, not vibes.
2. Cap review cycles — e.g. two human review passes per milestone included; extra is billable.
3. Forbid silent merges — agent may open PRs; you (or the client’s reviewer) merge.
4. Separate secrets — no production keys in agent-accessible env files the Project can read casually.
5. Plan limits — Cursor’s self-serve ladder (as of 21 Sep 2026 on cursor.com/pricing): Hobby free; Individual from $20/mo (Pro / Pro+ / Ultra variants); Teams $40/user/mo; Enterprise custom. Included usage + on-demand overages apply — re-check the live page before you quote a month of agent-heavy work.
6. Time-zone packaging only — overnight cloud runs can help async US/EU clients while you sleep; that is logistics, not an earnings promise.

If Projects burns your included usage by Wednesday, the SOW was underpriced — or the Project was too chatty. Tighten instructions; raise the fee; or split milestones.

Buy/skip sheet: stay in chat / start one Project / skip

Risk notes — CI auto-fixes, secrets, unreviewed merges

• CI auto-fixes — useful; still read the diff before merge.
• Secrets — rotate anything that ever landed in a chat log; keep .env out of agent reach when you can.
• Subscriptions to Slack / PRs — powerful; also a way to spend usage while you sleep. Start narrow.
• Beta — expect UI and limit changes; do not promise a client a frozen Projects UX for six months.
• Never let an agent merge unreviewed code to main on a paid client repo.

This week — open one Project for a real multi-file task you’ve been delaying

Pick one delayed multi-file task (not a typo). Create a Project. Paste the acceptance checks. Let the coordinator propose a plan. Approve or cut scope before subagents fan out. Review every PR. Write three lines in your notes: what context survived overnight, what you had to correct, whether chat would have been enough.

If the Project burned half your monthly included usage on planning chatter, shrink the instructions, disable noisy subscriptions, and re-quote the next milestone with review time priced in. Projects is a shipping aid — not a blank cheque against your Cursor plan.

Sources: Cursor Projects blog (10 Sep 2026), Cursor docs, Cursor pricing — re-check plan names and limits at paste time.

Soft next step

If you want the Project vs chat scope checklist on one page (~$2–3 / R30–50), that’s the printable sheet. Everything above stays free.

Get the 1-page Project vs chat scope checklist

Optional one-page printable of this free guide. Not a guarantee of clients, savings, or income.