Your Marketing Stack Was Built for Humans. AI Agents Ignore It.

01 /The Concept

Manicule is an AI agency that helps developer-tool companies create and maintain documentation for their products.

"AI agency" means the work is done by a combination of human specialists and AI agents. In practice, humans handle client communication and problem definition, AI agents do the heavy lifting, and humans then verify and refine the output before delivery.

Good documentation serves three concrete goals:

- Improving a product's visibility in AI chat tools like ChatGPT – so those systems recommend it more often when users ask relevant questions.

- Reducing the volume of support requests – lowering operational costs for the developer.

- Driving more sales – because developers who can quickly understand a product and evaluate its fit are more likely to buy it.

Manicule's team positions itself as a strategic partner, not just a writing shop. They're willing to think alongside clients about how to position a product and how to turn documentation into a genuine marketing asset.

The core insight here is that developer-users read the docs first. Documentation is the final marketing touchpoint before a purchase decision is made – it's where impressions harden into yes or no.

Manicule takes on the entire scope: information architecture, writing, illustrations, and ongoing updates.

A standard documentation build takes 15 days and breaks down into clear phases:

- Audit of existing docs and source code (days 1–2). A diagnosis of the current state.

- Documentation structure design (days 2–3). Building the table of contents and sidebar – the first thing developers see, and the first impression they form.

- Writing (days 3–10). Delivering polished concepts, quickstarts, guides, and API references.

- Code verification (days 8–12). Testing every code sample for correctness and functionality.

- Visuals and AI optimization (days 10–14). Creating diagrams, flowcharts, screenshots, and explainer videos, plus checking that the content is well-structured for AI chat systems.

- Launch (day 15). Link checking, cross-reference validation, responsive display testing, and sitemap submission to search engines.

Manicule is currently going through Y Combinator and has already signed its first clients from the YC alumni network.

02 /Why It Matters

Let's start with a bold claim that Manicule itself has never quite stated out loud. Every software product is about to become a developer product. That is, the primary users of most software will be developers – just not necessarily human ones.

As the founder of Box recently wrote, every employee in every organization will use AI agents in their work. As a consequence, the number of AI agents in a company will soon be 100 or even 1,000 times the number of human employees. And each of those AI agents will be using software to do its job. The number of "artificial" software users will exceed the number of human users by orders of magnitude.

At some point, AI agents will start choosing which software to use – based on what they find most convenient and effective for the task at hand.

And how will they figure that out? From the documentation, of course. Humans can rely on intuition and trial and error – AI agents will strongly prefer to read the docs first.

The natural interface for software used by AI agents is the API – calling functions is far simpler and more reliable than driving a browser to mimic human interactions in a human-designed UI.

That means "ordinary" software products are becoming API products. Which is exactly what a developer product looks like. Whether the product also has a traditional UI for human users becomes a secondary concern.

So documentation for developer products suddenly becomes a strategically critical question. The right answer could determine whether a product gets used by millions of AI agents – or ignored entirely. And the right answer is, of course, good documentation.

Interestingly, Manicule's own YC announcement already gestures toward this: they describe building documentation "for agents and humans" – with "for agents" in bold.

AgentDiscuss ([related review](/review/esli-ljudi-zahotjat-ty-opozdal)) also recognized that AI agents will soon choose their own tools. They launched a forum where AI agents can discuss which products work best for which tasks.

They've since evolved that idea into AgentRouter – a unified API for calling different products to solve similar tasks, with pay-per-call pricing. An AI agent can now not only choose the right tool but instantly use it.

On the documentation side, Waldium ([related review](/review/vot-kak-nuzhno-delat-mashinki)), another YC graduate, launched a platform for automatic developer blogging. Its AI engine monitors code changes in GitHub repositories and writes blog posts about updates, new features, and usage patterns.

Waldium has since evolved into an "AI-native CMS" with user call capabilities, support channel analysis, market research, and more. All of it serves the same underlying goal: promoting developer products – just with more data sources and output formats.

The key distinction between Manicule and Waldium is that Waldium sells a tool; Manicule delivers an outcome. Clients pay for finished documentation, not access to a platform. That reflects a broader trend – clients don't want instruments, they want results. And they're willing to pay more for results, which makes the AI agency model more financially attractive than selling raw software.

03 /Opportunities

The most direct opportunity from this review: build an AI agency for developer documentation, modeled on Manicule. The goal – making docs a marketing asset that works not just for human readers, but for AI agents.

But that points toward a broader opportunity: building all kinds of products and marketing tools aimed at an AI agent audience. Many tasks currently handled by people will soon be handled by agents – and agents need interfaces designed for them, and need to be reached through entirely different channels.

Put simply, a slick demo video is wasted on an AI agent. What it needs is a well-structured, comprehensive doc it can actually read. For humans, it's the opposite.

Consider Financial News Systems ([related review](/review/dva-poka-eshhjo-nelegalnyh-rynka)), which raised €1.5 million to build a news feed for professional investors where stories appear milliseconds after events occur – because the AI writes them instantly.

But millisecond delivery only matters if investment decisions are also being made in milliseconds. Which means the primary readers of such a feed can't be humans – they have to be AI agents capable of analyzing and acting at that speed.

That connects naturally to Marx ([related review](/review/kak-ne-poterjat-dengi-slushaja-sovety)), whose platform combines a live news feed with an AI agent discussion forum – so agents can reason through the news together and either make investment calls autonomously or surface recommendations for human review. Pair the two startups' products and you have a complete AI-native investment workflow.

So the big question: in which domains will AI agents become the primary consumers of software or information products? How does a product built for AI agents fundamentally differ from one built for humans? What can you add that would be redundant for people but valuable for agents? What can you strip out? How do you market to AI agents in the first place?

The answer space is enormous – but the most immediate entry point is picking a single professional software category where human users still dominate, then asking what that product would look like rebuilt for an AI agent as the primary user. Start there.