THE THING WE BUILT:

Deck Agent

Last week we shared our Deck Builder, a little app that lets you curate a slide deck solely based on the portfolio of work already living on our website.

That solved where the deck comes from. And it helped you solve which projects belong in it: you pick them, set the order, send the link.

But it exposed a bigger opportunity.

When we write a proposal, we’re already describing the client, their goals, their challenges, and the work we’re recommending.

That document already contains enough information to make the next decision.

So why were we making it twice?

The jump from manually assembling decks in Figma to pulling them from our portfolio database was a big step. But once we had those systems in place, the more interesting question became:

What other business decisions could we automate using the information we already have?

So we built a Deck Agent.

Deck Agent in Action

How It Works

The Agent tab lives inside the same Deck Builder admin. Same API. Same URL language. Same slideshow on the other end.

You give it a brief either:

A proposal from our proposal builder (pick from the list proposals you’ve already written), or

Pasted copy website text, mission statement, RFP excerpt, org about page, whatever you have

Hit Assemble deck. The agent:

1. Understands the org name, sector, mission, what kind of engagement Township might be pitching, what proof themes matter, what to avoid

2. Compares every project in the portfolio public and private and scores each one strong, moderate, weak, or none, with a reason

3. Picks a deck project slot codes in order, same /0103 URL format the builder already uses

4. Hands off to you You can open the link if it’s ready to go, or Edit in Builder to tweak the order or add to it

Notes from Building

1. The proposal builder was already in the stack.

The Deck Agent was a quick build because we already invested in structuring our data.

Our proposals aren’t a loose arrangement of PDFs, they’re consistent markdown with a pretty full client context and metadata that’s easy for software to read. That wasn't a trivial investment of time, but it’s the foundation that makes workflows like this possible (and way less painful).

2. The agent is a draft, not a verdict (a human in the loop)

Like all good AI workflows, the agent creates a strong first draft the human makes the final call.

You can uncheck projects, reorder the deck, add or remove work, and save the result. The goal isn’t to replace judgment; it’s to eliminate the repetitive work of assembling a solid first pass. Human taste still wins on the final send.

3. Integration over replacement (again).

We didn’t build a new portfolio brain. We added a tab, a server action, and a new LLM API call that reads the Work API catalog and proposal store we already had. A bite sized additional lift.

4. The Bigger Picture

The Deck Agent isn't really about decks.

It’s about recognizing that your business already produces structured knowledge.

Every proposal contains information about the client, the industry, the goals, the deliverables, and the proof points that matter. Most companies create documents like this every day and then treat them as static files that get emailed away.

Here we’ve started asking a different question:

What else could this document do?

For us, the answer this time was “assemble a portfolio presentation.”

For someone else, it might generate a project kickoff, recommend a project team, populate a CRM, draft a discovery agenda, or flag implementation risks.

The pattern is the interesting part.

Once your business information lives in systems instead of scattered documents, you can start building small agents that connect one workflow to the next.

This one just happens to build presentations. What’s next?

Built with ❤️ by Township

You’re receiving this because you signed up for Township’s “We Built a Thing” newsletter. We'll only send these when we, well, build things.