Esmero

Allowing creatives to iterate faster

Comment

Inspiration

love to wonder, tell stories, experience and create.

I'm constantly writing on Notion, brainstorming on ChatGPT, drawing on my notepad and whiteboard, designing on Canva/Excalidraw and building with Claude Code, on top of an IDE or engine.

Allowing my unconscious to drift out brings me great joy. I can spend my life creating. I want to create faster. To be able to externalise with less friction.

I have spent the past year building AI agents. I am convicted that this technology is underrated.

The more one iterates, the closer one gets to the "truth" in ones domain.

Without my voice and hands, I won't be able to create. Without my eyes, nose and ears I won't be able to iterate.

The single most important metric for these companies should be time to iteration. Empowering the hands and voice for creation and providing the eyes and ears with quality feedback for iteration.

Gen AI completely changes the way to build creative tools.

Miro, Canva & Notion for example believe that giving more tools will empower users. That "AI" is just one of these "tools". New users encounter too much friction to learn to use these platforms since they are overfitted and complex. They are currently fundamentally wrong in the way they build creative tools and are outdated to leverage AI as a foundation.

The creative needs to iterate as fast as possible, for thats what brings one clarity. The metric is time to iterate. Minimising time to iterate means working with fundamentals to build a product that has minimal tools which will allow the user to externalise their thoughts as fast as possible and receive quality output.

Now is the time to reimagine creative products, leveraging AI as it's foundation while others just "plug it in".

What it does

Esmero is a writing canvas with an intelligent agent built from the foundation. The Canvas: Where you write. Clean. Focused. No distractions.

The Agent: Understands context at every layer. It sees: Your conversation history, what you've discussed Your current writing, the live canvas Your selection, what you've highlighted Your project's documents,related work you've created

When documents exceed 5, the agent uses LLM-based relevance filtering to intelligently surface only what matters to your current message. No noise. Just signal. Intelligence in Action: You ask anything. The agent responds in conversation, or suggests edits directly to your canvas. Edits appear as diffs, red for deletions, green for additions. Accept what resonates. Reject what doesn't. Each interaction is contextual. Each response is truthful. The agent doesn't pretend to know what it doesn't. It works with what exists. No complexity. No friction. Just expression and intelligence, working together.

How we built it

From scratch, created repo this morning. Started with the canvas. Built the writing interface, the sidebars, the project structure. Frontend first. Then the backend. Express server, agent architecture, context orchestration. We ran tasks in parallel, frontend dev server, backend dev server, multiple implementation streams simultaneously. he agent system follows a layered intelligence pattern: Context determination - Parallel execution gathering conversation + file context Relevance filtering - LLM-based selection when documents exceed threshold Prompt construction - Truth-based system prompts, no prescriptions Structured responses - Zod schemas for reliable diff generation

Built with Claude Code for execution, used canva and google studio for logo. Wisperflow for continuous voice direction.

Challenges we ran into

Front end.... Spent too much time building the fe... Damit. Making things work beautifully and bugless is truly a work of patience. Still many slight bugs, but not an issue for the state of things.

Accomplishments that we're proud of

Building a full working agent with utility in lightning speed. Still a lot of iteration to be done, but glad it works.. 100% built today. From logos, to code to open ai keys etc etc...

What we learned

In hackathons, dont spend too much time on front end... Didn't allow me to use any of our partners, which could have been fun to explore our agent with... Voice + claude code is absurd... Had many instances running and spent 80% of my time just speaking. None stop iteration... What a time to be alive.

What's next for Esmero

Hopefully more iteration and more building, at a general level. If I am still feeling joy after a while, will probably raise a small round to go full time. I truly believe iteration is what we should aim for if we are creating. The more we iterate the faster we get to truth. Be this in writing, video making, coding etc... Esmero could allow others to iterate faster, and allow creatives to excel in this world of AI.

Built With

Share this project:

Updates