From "no idea" to a real business in one afternoon

Replit Free Mode

Thinking used to cost you. Now it doesn't. 🧠

Here's the thing nobody says about AI builders: the reason people ship bad products isn't bad taste, it's that every message felt like it cost something. So you skip the brainstorm. You skip "what are the ten versions of this," you skip "who else is doing this and where are they weak," and you go straight to building the first idea you had — which is almost never the good one.

Free Mode flips the economics. An efficient, fast model handles the thinking — ideation, research, analysis, planning, documents — without eating your credits, and the expensive models stay reserved for the work that genuinely needs them. Replit's agent tells you when you've hit that line and asks before switching.

Which changes the shape of the workflow entirely:

• 💭 Think wide, free. Ten ideas instead of one. Kill nine.
• 🔎 Research hard, free. Competitors, their weaknesses, your positioning, your price.
• 📄 Document everything, free. GTM report, prospect list, outreach drafts.
• 🔨 Spend credits only on the build. The one part that actually needs the heavy model.

What you need — and what "free" actually means ✅

Let's be precise, because the name is doing a lot of work: Free Mode means the work doesn't burn your credits — not that Replit is free. Starter accounts include daily Free Mode usage, which is enough to run the thinking stages and see whether this workflow suits you. The real limits sit on Core (about US$20–25/mo) and Pro, which is where an afternoon like mine lives comfortably.

At launch, Replit published Free Mode limits per cycle (resetting roughly every five hours): up to 30 hours of chat, 60 projects, 1,000 designs, 200 slide decks — and claims up to 30x more creation on the same subscription. Generous, not infinite; check what's current when you read this.

Also useful: the UI shipped alongside it puts asking, planning, and building in one place, which is what makes a chain like the one below feel like a single conversation instead of six tools.

Start vague. On purpose. 📋

This is the actual prompt I opened with. It is deliberately, almost embarrassingly unspecific:

I want to sell something to local business. no idea what. brainstorm me

That's it. No spec, no plan, just a direction. Most people would be too embarrassed — or too cost-conscious — to send that. Free Mode is exactly what makes it sendable, and vagueness is a feature here: you want the model's whole spread before you narrow, not a polite version of the one idea you already had.

If you want more shape, use this:

I want to sell something to [local businesses / a market you know].No fixed idea yet. Give me 10 distinct product concepts.For each: who exactly buys it, the painful problem it removes, roughlywhat they'd pay, and how hard it is to build (1-5).Range from boring-and-obvious to genuinely weird. Do not converge.

Kill nine. Interrogate the survivor. 🔪

Ten ideas came back. I killed nine and reshaped the one I liked. The killing is the work — and the model is better at helping when you make it argue against itself:

I'm keeping #[x]. Kill it for me first: give me the 5 strongestreasons this fails — market, competition, distribution, cost, and theone nobody thinks about.Then, if it survives: reshape it. What's the smallest version thatsolves the whole problem for ONE specific type of customer? Name thatcustomer precisely.

Who's already doing this, and where are they weak 🔎

Now the research chain. This is the part that used to feel expensive and now doesn't:

For [the reshaped idea], research the market:1. Who's already selling this to [customer]? Name real companies with   links — not archetypes.2. For each: what they do well, and specifically where they're weak   (pricing, onboarding, support, features, who they ignore).3. Where do I position against them? Give me the one-sentence wedge.4. What should I charge, and why — anchored to what they charge.Mark any number you could not verify from a source as [VERIFY].

Then the document:

Write me a go-to-market report I could hand an investor: market andcustomer summary, competitor table, my positioning and pricing,channels ranked by how fast they'd work for a solo founder, and a90-day plan with weekly milestones. Keep every claim sourced.

Now build — and let it tell you when to switch modes 🔨

Only now, with a validated idea and a written plan, do you spend the expensive credits:

Build [product] for [specific customer], based on the plan above.Start with the smallest working version: [core feature], nothing else.Design direction: [clean/bold/playful] — match the positioning we set.Tell me before switching to a heavier mode, and why it's needed.

That last line is the discipline. The agent proactively suggests stepping up when a task genuinely needs more — you want that prompt to arrive as a question you approve, not a surprise on your usage. When it asks, ask yourself whether the build actually needs the heavy model, or whether you're just impatient.

Then the rest of the asset stack — all thinking work, not heavy building:

From the same plan, create:- a pitch deck for my first prospect type (short, one idea per slide)- social assets announcing the launch, in the positioning's voice- a prospect spreadsheet: company, contact, role, why they fit, source- a first-touch outreach email written off what you know about each  company — save as drafts, do not send anything.

🔑 What makes the afternoon work

1. 1Think wide before you narrow. Ten ideas, then nine funerals.
2. 2Make the model attack your idea before it builds it.
3. 3Research before building. Building the wrong thing fast is still wrong.
4. 4[VERIFY] flags get sourced or cut — always before anything reaches a prospect.
5. 5Approve every mode switch consciously. Free for thinking, credits for building.
6. 6Drafts first. Nothing sends from your inbox unreviewed.

The honest notes ⚠️

• "Free Mode" is about credits, not price. It means the everyday thinking work doesn't consume your usage — not that Replit costs nothing. Starter accounts get limited daily Free Mode usage; the real headroom is on Core and Pro. Anyone telling you AI building just went free is skipping a step.
• It has ceilings, and they reset. Published limits at launch ran per five-hour cycle. You'll feel it if you go all day, and the answer is pacing, not panic.
• The heavy work still costs. Free Mode covers ideation, research, analysis and documents. Real builds step up to the paid modes — that's the deal, and honestly it's the right split.
• A hundred AI-found leads is a starting point, not a list. Some rows will be stale, wrong-role, or plain invented. Spot-check before you contact anyone, cut ruthlessly, and send small personalised batches — check the rules where you and your prospects are (PDPA here in Singapore, CAN-SPAM in the US, GDPR in Europe). Twenty real prospects beat a hundred guesses.
• My own read, six months in: the new UI is a genuine improvement, responses are noticeably faster, and I'm getting meaningfully more out of the same monthly spend. That's my experience, not a benchmark — Replit's own claim is up to 30x more creation on the same subscription, which is a different number measured a different way. Judge it on your own account.
• One afternoon ≠ one afternoon to revenue. I built the whole stack in an afternoon. Selling it is a separate job that takes longer than the building ever will — that's the actual lesson of the last two years.

Links 🔗