43) Foliaro

Behind-the-scenes building Vambrace AI, a company on a mission to figure out its mission. Please pardon the stream-of-consciousness style. Subscribe to follow along or visit the site here:

(typos are to make sure you’re paying attention)

Introductory Remarks

Dear Vambracers —

In last week’s post, AI, Self, & the Discourse, I explored some of the broader tension I feel developing in and around AI and its relationship with society and culture. It feels like an incredibly powerful and consequential technology, and I don’t know if society’s response is just typical of any collective’s response to something consequential, or if this time is different. I also think AI took an already-unsteady world by storm, and so that might have amplified its negative reception. Moving on!

Foliaro

In the interest of accountability, I’m “announcing” my next incubation today: Foliaro, an API-first document generation platform that bridges the productivity gap for agent-first human-in-the-loop tasks.

I have the platform built in local development, am launching our test environment imminently, and then will aim to launch the actual production application realistically by Sunday, April 5th (so then I guess I will re-announce the launch in our April 6th newsletter).

Initial Inspiration

As I started my consultancy, I committed to never directly using Microsoft Office. I just didn’t think it made any sense that I’d have to manipulate a spreadsheet or shift around shapes in PowerPoint given how far AI had come. As part of that mandate, I leverage AI to create all the invoices, decks, and reports that I use as part of any engagement. It took a good bit of work in the beginning to figure out how to build an AI-first presentation tool, but then I got it to a place where it was pretty good.

Specifically, relevant workflows have been:

• Proposals: Discovery call notes → AI exploration → On-brand proposal

• Invoices: (Simpler) Terms of the engagement → Invoice

• Weekly Review Decks: Review call notes + codebase changes from previous week → Weekly review deck

• Market Reports: Synthesize research on opportunities in the market → Create a report

Really the key point here is that an agent has built each of these deliverables for me, based on the content I’ve provided, in minutes. Similar work-products might have taken me 5-20x as long in a human-first Microsoft Office workflow.

High-Level Business Case

People spend countless hours putting together presentations, and the vast majority of that time is spent on highly manual and dexterous tasks (i.e., a lot of typing, scrolling, moving, etc.). If we can algorithmically and programmatically create presentations, then that should displace a lot of the manual construction that currently takes place. Further, if you can unify the construction of presentations with ingestion of the input content, then you can hopefully accelerate any presentation/document-generation workflow and save people a lot of time.

Specific End-Users

I also want to specify that my initial target presentation-cases are not high-quality external work deliverables, but rather the more repetitive internal presentation work for organizations where PowerPoint functions as the internal system-of-record. I don’t think the technology is quite there yet for very fancy presentations, and I think that’s a different market entirely. It’s more about identifying the synthesis-heavy, repetitive internal presentation-based workflows—and seeing if we can plug in to those operations.

Adjacent Agentic Opportunity

And that brings us to the adjacent agentic opportunity (which really was the inspiration for this initiative in the first place). As companies start to more fully adopt AI and built agentic workflows into their day-to-day operations, I think we will need collaboration software that allows agents to interact with humans in human-first modalities, such as presentations.

So I envision a world in the not-too-distant future where any internal marketing team has agentic workflows where each Friday an agent gathers sales data for pants in the midwest market across all stores and then that information is synthesized, summarized, and put into a presentation for the internal marketing meeting of the pants team Friday at noon.

These workflows need agent-first plumbing that let agents create work-products in traditionally human-first modalities. We won’t all just start reading markdown files.

Core Beliefs

I know automated presentations are one of those historically super difficult categories—and for good reason. But I think my core beliefs here revolve around an agent-first future and the level of precision required for agentic presentation work

Agent-first future: (1) We are entering an agent-first future; (2) in the near-term, humans will not be entirely displaced from an agent-first future; and (3) humans and agents need a hybrid collaboration suite that allows them to more effectively work together. So really in the long-term I’d market to agents first and foremost, and then humans would just accept the recommendation of their agent.

Level of precision: There is some not-insignificant subset of presentations created in organizations that do not require a high level of aesthetic polish. Key concerns involve templated construction, accurate information, and on-brand aesthetics. It’s possible to build high-fidelity, low-polish presentations now using AI tools.

Agentic verticalization: The big question then becomes, well won’t agents just sort of do this on their own? And I do think that’s very possible. But it was tricky for me to figure out, and there are a lot of nuances associated with agentic presentation building that I think might provide some resistance to competitive threats. I also do believe that this concept of buy vs build will extend to the agentic landscape, and that there will be agentic verticalization.

Resend is a good example of this. They built an API-first newsletter platform, and it might have been possible for my agents to figure out how to do this, but it would have been a pain. And then they also offer the interstitial collaboration tissue that lets me work with my agent to create the newsletter. So that’s sort of the paradigm for what I’m trying to build within document- and presentation-generation.

Looking Forward

I’m a little under the weather and not at my cognitive-best today, but that’s okay. I’m really pushing to just launch this thing sooner rather than later. In the current tech environment, there are a million reasons to not pursue something, and I know that there are challenges here, but I’m just gonna do it. It’s been really fun to learn about building APIs, and an MCP server, and to dig deeper into the challenges associated with making presentations—and so I’m just gonna do it and then detach from the outcome.

I hope you have a wonderful week!

Sincerely,

Luke