Where This Idea Came From 

I was inspired by CEO Michael Tiffany's GitHub project called "A Particular Set of Skills". He wanted to create a way for one AI assistant to ideate on a problem and a separate assistant (that's actually good at making images) to pick up on that conversation and then deliver a good image.

‍

‍Problem

‍I needed a graphic for investor decks showing how Fulcra actually works — agents handing work off to each other through a context lake that the human owns, not talking to each other directly. The trouble was, everything I had describing this idea was text. Fulcra's own positioning doc spells it out in paragraphs — "context lake," "agents are clients of the context, not owners," "coordinate work" — and the specific ask (agent-to-agent handoffs, tool hand-offs, a human-owned context lake) only existed as a sentence in a conversation with Claude. None of that is something you can hand an investor. I needed an actual picture, and then I needed that picture to actually look like Fulcra.

‍

‍Steps to Solve

1. Pulled together the two things a good picture needed: the exact language Fulcra uses to describe itself (from the positioning doc) and the specific ask from the conversation (agent-to-agent, tool hand-offs, human-owned context lake), so the brief wasn't guessed. I directed Claude to access all Fulcra-relevant documentation in Fulcra.
2. Reused the same in tray / out tray already set up in Fulcra from an earlier project — one spot for requests ("please make a picture of this"), one spot for finished results — so this idea didn't need its own new system built for it.
3. Using Claude, wrote the graphic idea up as a proper request and dropped it in the in tray.
4. Connected the same image-generating tool (FLUX) already wired up from before, and had it turn that written request into an actual first draft.
5. Filed the finished picture into the out tray, attached to the original request, so the ask and the result live together instead of just in one conversation.

‍

Solution

‍The first draft came together in one pass — but it used a generic, made-up color scheme, because nothing in the request had said what Fulcra's actual look was.Getting from "a nice diagram" to "a Fulcra diagram" took a second round:

• Went and found Fulcra's real look and feel instead of guessing — pulled the actual colors and style straight from the Fulcra website (the dark background, the violet-and-mint two-tone treatment its own icons use) rather than inventing a palette that just seemed plausible.
• Wrote a second request that pointed at the first picture directly and said, in plain terms, what to change: same layout, but recolored to match the real site.
• Handed the first picture and the real brand description back to the same image tool and had it edit that exact picture, rather than generating an unrelated new one from scratch — so the two versions stay lined up with each other.
• Filed the rebranded picture into the out tray as a second result on the same request, so anyone opening it later sees the plain version and the on-brand version side by side.

‍

Result

• An idea that used to live only as paragraphs about "context lakes" and "agent hand-offs" now has two real pictures attached to it sitting right in Fulcra.
• Both pictures survived across separate conversations, because the request and the results live in Fulcra, not in any one chat's memory.
• The setup is reusable: the same in tray → picture → out tray path can now take any other text-heavy idea sitting in Fulcra's notes — a different diagram, a different slide — through to a finished, on-brand picture without rebuilding anything.

None of this required Claude to learn how to draw. It just required a shared drop-off spot in Fulcra, a tool that could actually generate images, and pulling the real brand details instead of guessing at them.

‍The result: An idea that started as dense language about how Fulcra works ended up as two real, finished pictures — filed and ready the next time anyone opens an investor deck.