Author: shelly@maybe.ai
The Problem Everyone Misses
Everyone's building AI tools. Chatbots. Agents. Workflow builders.
They assume users know what they want. Then compete on who executes better.
But what we see: Users don't know what they want.
A client came to us. Said he wanted to "automate competitor monitoring."
We asked: Which competitors? He said: The main ones in the industry.
We asked: What data? He said: Pricing, promotions, new products.
We asked: Where to scrape from? He got stuck.
We asked: When to alert? He thought for a while.
We asked: What if scraping fails? He said: Um… should we retry?
Execution Isn't the Problem
This request, technically, isn't hard. Write a scraper. Run a cron job. Send notifications.
The hard part: turning "automate competitor monitoring" into executable specs.
Which URLs. Which CSS selectors. What frequency. What counts as success. What counts as failure. How many retries. Retry interval.
These all need clarity.
Without clarity, tools don't matter.
LLMs Are Misunderstood
Everyone says: LLMs let non-coders build apps.
Wrong.
LLMs don't think for you. LLMs help you decompose your thinking.
They're good at planning:
- Understanding vague intent
- Breaking down into steps
- Matching appropriate tools They're bad at execution. Too expensive. Unstable. Unpredictable.
Our Approach
Step 1: Plan with LLM
Dialogue. Ask questions. Force clarity.
This is what LLMs do best. Flexible. Understands natural language. Probes for details.
Step 2: Execute with Fixed Code
After planning, don't run with LLM. Generate fixed code workflow.
Code is stable. Predictable. Cheap. Runs for cents, not dollars.
Use dataframe for data processing. Use loops for retry logic. Use if-else for branching.
These don't need LLM intelligence. They need determinism.
Step 3: Give It Hands
Brain alone isn't enough.
LLM can say "scrape competitor.com for prices." But it can't actually scrape.
Our human-like browser automation:
- Opens web pages
- Clicks buttons
- Fills forms
- Waits for page load
- Handles popups Manual work. But if AI replaces human repetitive work, it needs to do manual work.
Why Now
Three timings converge:
LLMs are smart enough. Can understand complex intent. Can probe details. Can plan steps.
Code generation is cheap enough. From dialogue to workflow, cost is accep
table.
Browser automation is mature enough. Playwright, Puppeteer are stable. We just need to teach LLMs to use them.
We're Not Another Workflow Builder
We're a problem decomposition tool. That happens to generate workflows.
You think you need execution tools. You actually need thinking tools.
With clarity, execution is easy. Without clarity, no tool helps.
MaybeAI: From "I want this" to "This is exactly what I want."
That's what we're building.
About MaybeAI
MaybeAI is a business data workflow automation platform that lets prosumers describe their data needs in natural language and automatically handles the complete "acquire → analyze → act" business cycle, with intelligent solutions that learn and evolve with each use.
Data workflows, minus the work.
Top comments (0)