Imperial Labs

What it is

Imperial Labs was going to be an AI marketer. Not a chatbot you ask for tips — a system you hand a project to, that then tells you what to promote, where to promote it, and how, and generates the actual material on a schedule. The only thing you were supposed to do is describe your project once, in a brief that's already half-filled in.

That's the version that lived in my head. This case study is honest about the gap between it and what exists: the ambitious AI platform is shelved, and the pieces that actually shipped are a handful of small, sharp marketing tools — the kind of thing that survives when the big idea doesn't. Those tools are embedded live further down; you can use them right here on the page.

The problem

If you build things alone, marketing is a second full-time job you're not qualified for. You finish the product and hit a wall that has nothing to do with code: what do I even say about this, and where do I say it? Which channels. Which angle. How often. What a week of posts should look like. Every one of those is a decision, and a solo builder makes all of them badly, late, or not at all.

The existing answer is a drawer full of disconnected apps — one for scheduling, one for UTM links, one for ad copy, one for meta tags, a document somewhere with a half-written "strategy." Nothing knows about your project, so you re-explain it to every tool. I wanted the opposite: one place that knows the project, and does the thinking.

How it was meant to work

The whole system hangs off one object: the brief. You fill it in once per project — and it's not a blank page, it's a structured form with the fields that actually shape marketing decisions:

• business name and product description

• target audience and countries

• goals and tone of voice

• competitors and constraints

From that brief, the AI was meant to run a chain — and this part was really built, the pipeline exists and produced output:

1. Strategy — the "what and where." Given the brief, it proposes the channels worth being on and the angle for each. This is the answer to "I don't know where to promote."

2. Campaign Plan — the "how." Pick a goal and a landing URL, and it lays out a concrete campaign: phases, messaging, what each channel is doing.

3. Content Calendar — the "when." It turns the plan into a dated schedule of posts.

4. Content Pack — the "actual words." It drafts the posts themselves, per platform, using a library of 30 content templates (launch-day announcement, feature spotlight, limited promo, customer story, event reminder, hiring call, and so on).

The step I was proudest of designing: the output didn't just sit in a chat window. It applied itself to the workspace. One click turned a generated plan into real tasks; another turned a content calendar into real dated events. The AI's answer became your to-do list and your schedule, not a wall of text you still had to action. That was the whole thesis — describe the project, get a working plan you can execute, not advice you have to interpret.

Here's the actual interface — the real thing, re-themed dark and frozen into a clickable demo. Walk it the way you'd walk the original: open the project, generate the strategy, turn it into a campaign plan, apply it to the workspace. The AI is switched off, so every click just reveals the result the working system already produced — no API is called, nothing is spent. Expand it for room to click around.

Around that core sat the everyday marketing kit: campaigns, a UTM builder with history and shareable links, an ad-copy checker, a meta-tag pack builder, SEO helpers, a template library, and a workspace to collect it all.

What actually shipped

Here's the honest line between vision and reality. The AI chain worked in the sense that it ran and produced plausible output — but "produces plausible output" and "produces output a real business should act on" are very different bars, and closing that gap is most of the work (more on that below). What I'm willing to stand behind, and what I kept alive, are the deterministic tools — no AI, no API bill, no judgment calls. They do one small thing correctly, every time.

I ported them out of the platform and onto this site, rebuilt as pure browser tools. They're embedded here, live. Nothing you type in them is sent anywhere.

Config Redactor

Not a marketing tool — a developer one — but it's the piece I reach for most, so it shipped first. Paste a log, a config, or a whole .env, and it masks every secret (keys, tokens, passwords, private keys, AWS/GitHub/Slack credentials) so you can share it safely. Originally this was a small server-side app of mine called SafePaste; here it's rebuilt to run entirely in your browser.

Ad Copy Fit

The one that came straight from the marketing platform. Write your copy once and see, at a glance, whether it fits every platform's character limit — X, LinkedIn, Instagram caption and bio, YouTube, Google Ads. Green means comfortable, amber means it's getting long, red means over. The limits are the real ones.

UTM Wizard

Tag a campaign link so analytics can tell you where traffic came from. Pick a channel preset, fill the campaign, copy a clean URL — existing query parameters are preserved, empty fields are dropped.

Meta Tags Generator

Fill in a title, description and image once, and copy a full set of SEO, Open Graph and Twitter card tags — with length hints that keep your title and description inside what search results actually show.

Meta Pack Builder — described, not embedded

One tool I'm deliberately not putting live yet, because it's the one with real nuances to work through. Meta Pack was the "do the whole head of the page for me" tool: give it the basics and it assembles the complete meta bundle — SEO tags, Open Graph, Twitter cards, and the supporting bits — as a single downloadable file you drop into a project. Unlike the others, it was built server-side and it produces a file, not just on-screen text, so porting it to a pure browser tool means deciding how the download works, what's in the bundle by default, and how far it should go (favicons? a web manifest?). It's a real tool with real decisions attached — worth doing properly rather than fast. It gets its own pass.

Why I shelved the platform

The tools were the easy 20%. The AI marketer was the 80%, and it's where I ran out of road.

The generation ran. What it couldn't do was be reliably right. A marketing strategy that's 80% plausible is worse than useless — it's confidently wrong in ways a non-expert (me) can't catch, and acting on it costs real money and real audience. Making the output trustworthy wasn't a prompt away; it needed evaluation, guardrails, a feedback loop with actual campaign results, and honestly a marketer's judgment baked into the system. Every one of those is its own project.

On top of that: the platform lit up an API on every meaningful click. A product whose core loop is "press button, spend money on tokens" needs either paying users or a very deep pocket, and I had a hypothesis, not either. The same lesson I learned building SENU — pivot early, it's cheap; pivot late, it's expensive — applied here, and this time the honest pivot was to stop.

So I did the same thing I did with Didko when the paid-product route got too heavy: I kept what was real and let go of what wasn't. The AI platform is off. The tools are on.

Identity

The project carried my Imperial brand — the umbrella I put over the things I was building before they earned their own names. The plan was to launch it as Imperial Labs, on imperiallabs.io. I built the whole thing inside a separate codebase I called the Toolbox, meaning to graduate it onto the Labs domain once it was ready. It never got ready, so the domain stayed empty and the work stayed in the Toolbox. That's the small, unglamorous truth behind a lot of shelved products: the final address was reserved, and the moving van never came.

Lessons

A tool and a platform are different animals. A tool does one deterministic thing and you can finish it in an afternoon. A platform makes judgments, and judgments have to be right, which is a bottomless amount of work. I budgeted for a platform like it was a pile of tools. It wasn't.

"The AI produced something" is the start of the work, not the end. Same lesson as SENU, sharper here: the model generating a strategy is the easy part. Knowing whether the strategy is any good — and being accountable when it isn't — is the actual product, and I couldn't build that alone.

Ship the part that's true. The most honest outcome of an over-ambitious project isn't a graveyard — it's the working pieces you rescue from it. These four tools are genuinely useful, they cost nothing to run, and they'll outlive the platform that spawned them. That's not failure. That's the residue of aiming too high, and it's worth keeping.