Two proposals in one. First: hand the mechanical half of your production — the exports, the resizes, the renaming, the re-exports after every revision — to a machine that does it in seconds. Second: use the same tool to build the websites you're currently designing and giving away.
You already know this shape. The part worth staring at is the black box: every revision round re-runs the entire export matrix by hand, and no client has ever paid extra for round three.
Current pipeline
Email, call, or a Word doc with the sizes buried in a paragraph.
You translate vague requirements into a deliverables list, mostly from memory.
The actual work. The part you trained for and the part that's worth your rate.
Every size, every platform, every density, every colour space. By hand.
Naming conventions, folder trees, zipping, uploading.
Back to 03. Then all of 04 and 05 again. Usually twice.
Then the web build goes to a developer, and that invoice isn't yours.
Steps 04 and 05 are pure mechanism. There is no judgement in them, no craft, no thing a client could tell you did well. They're also the steps that scale worst — a second brand variant doubles them, a second language doubles them again.
The five apps this touches
Culling, batch develop, consistent colour before anything reaches Photoshop.
Retouch, composite, mockups, and every social and web image that ships.
Logos, icons, die-lines, anything that has to scale or go to a cutter.
Multi-page print, brand guidelines, decks, anything with a page count.
Proofs, preflight, PDF/X for the printer, client-facing markup.
A modest campaign for a single client. Nothing exotic — this is a normal week.
Forty-one exports is roughly two to three hours of clicking, and it happens three times per project. That's most of a working day per client, every client, spent on the least valuable thing you do — and it's the first thing that gets rushed when a deadline moves.
Read this carefully, because the boundary matters: nothing here makes a design decision. You design exactly as you do now. The automation starts the moment a design is approved and stops the moment it's delivered.
Layer A — the brief becomes a file
Paste the email, the call notes, the messy spec sheet. Whatever you actually received.
A clean, checkable list of every file the job requires — sizes, colour spaces, formats, names.
Thirty seconds of review. This becomes the contract and the export instructions at the same time.
This one step alone kills the two most expensive mistakes in the business: quoting for the wrong scope, and discovering on delivery day that they also needed a billboard.
Layer B — the exports run themselves
Your PSD, AI, or INDD. Layered, named, untouched.
Every social size, every breakpoint, every pixel density. Crop or fit chosen per deliverable, @1x/@2x/@3x, sRGB conversion, JPEG quality set once. Your master is duplicated and closed unsaved — the PSD you opened is the PSD you keep.
Every lockup × every colourway × every format, one pass. Primary, stacked and mark; full colour, mono and reverse; SVG, EPS, PDF and PNG — with the SVG cleaned up rather than dumped out full of Illustrator's metadata.
Variants driven from a spreadsheet. Ten SKUs, three languages, same effort as one — this one leans on data merge, which is a native InDesign feature, not something bolted on.
Bleed, marks, CMYK, the printer's named preset — checked to exist before the run starts, so it fails at the beginning with a clear reason rather than at the end with a bad file.
Runs on the Creative Cloud licence you already pay for. There is no Adobe API bill, no enterprise contract, no plugin to buy — this drives the apps through the same scripting doors Adobe has left open since the nineties.
Not all five apps automate equally
The honest read: Photoshop and InDesign carry this. Illustrator repays the effort because identity handoffs are so repetitive. Lightroom mostly stays manual, and that's fine — culling was never the part eating your week.
Layer C — delivery
Built to your naming convention, identically, every project.
A single contact-sheet PDF so the client reviews everything in one place.
Zipped, uploaded, link sent. Revision round two re-runs the whole thing in seconds.
These apps don't run headless. Your machine has to be on, logged in, with the app running — so this is a fast assistant sitting at your desk, not a server farm. In practice that's fine: you press go, it works through the matrix, you get coffee. But it's worth knowing the ceiling before anyone promises you the moon.
Two smaller ones, said now rather than discovered in week three. The first run has to talk macOS into letting one app drive another — a permission dialog, once, and it is the fiddliest fifteen minutes of the whole setup. And scripts address Adobe apps by exact version name, so an annual upgrade can stop them working. The fix is a one-line edit, but somebody has to know that's what happened.
This is the bigger number. Every time you design a website and hand a layout to a developer, you're doing the hard half of the job and invoicing for the small half. The reason you couldn't build it before was syntax — not judgement. Syntax is exactly what these tools are good at.
You invoice: the design fee.
You invoice: design + build + hosting.
Market rates vary, but a marketing-site build is typically one and a half to three times the design fee, and a small monthly retainer for updates and hosting turns one-off projects into recurring revenue. That's the whole pitch: the same client, the same project, a materially bigger invoice.
Say yes to this. Not to that.
The rule underneath: if it breaks at 2am, can you fix it? A static marketing site can't really break — it's files on a server. An app with logins and payments can, and that liability is worth more than the fee.
Both major tools are sold the same way: the coding agent is bundled into a chat subscription rather than priced separately. Claude Code comes with a Claude plan; Codex comes with a ChatGPT plan. Prices below are current as of August 2026 and worth re-checking — this market moves quarterly.
| Tier | Claude | ChatGPT / Codex | Who it's for |
|---|---|---|---|
| Entry | $20/mo $17 annual |
$20/mo | Everything in this document. Enough for daily export scripts and building a site over a few weeks. Start here, and only move up if you're actually hitting the wall. |
| Middle | $100/mo | $100/mo | Roughly five times the entry allowance. Worth it if you're waiting on limits several times a week — not before. |
| Heavy | $200/mo | $200/mo | Twenty times entry. Full-time agent work, several running at once. Almost certainly not you. |
| Adobe | $0 extra | Your existing Creative Cloud licence is all that's needed. The scripting route requires no Adobe developer account or enterprise contract. | |
| Hosting | $0–20/mo | Per client site, and billed to the client. Small sites often land in the free tier. | |
Both meter usage on a rolling five-hour window with a weekly cap on top, and that allowance is shared with your normal chat use — so a heavy afternoon of chatting eats into the evening's coding. Neither publishes exact message counts, because a long task costs more than a short one. The practical test is one month at $20. If you never hit a limit, you're done; that's your price.
The good reason: it can look at a rendered preview of what it just made and correct itself, which matters enormously when the output is visual. The better one is less technical — Adam already runs this daily and can unstick you in ten minutes instead of a weekend. Tool support beats tool specs.
Deliberately slow. Every week produces one finished thing you'd use again, and no week puts a client project at risk.
Subscribe at the $20 tier, install the desktop app, and pick the single export you do most often — probably your Instagram set or your standard web hero sizes. Get one script working end to end on a finished old project where nothing is at stake. Time yourself doing it manually first, so you have a real before-and-after number.
Take three old client briefs and turn each into a clean deliverables list. Refine the format until it matches how you actually quote. From now on this file drives both your estimate and your exports, so scope creep becomes visible instead of absorbed.
Every identity job ends with the same twelve-to-twenty-file handoff. Build it once, properly, with your naming convention baked in. This is the highest-value script you will write and it pays for the subscription in a single project.
The pilot with zero client risk. You know the content, you're the only approver, and the deadline is whenever. At the end you have a live site you built yourself — which is simultaneously the proof that you can do it and the portfolio piece that sells the service. Put "design and build" on your rate card the day it ships.
Not a discount to win it, and not a rate you'd resent. Price the build at what a developer would have charged, because a developer is no longer taking that money out of the room. Offer the hosting retainer as a small monthly line — most clients say yes because the alternative is finding someone else at 2am.