Inspiration
Companies make high-impact decisions based on assumptions:
“This is our biggest growth opportunity.”
“Customers want this.”
“Our competitors are moving in this direction.”
“This market is about to expand.”
The problem is that the evidence needed to validate those assumptions is scattered across the web, competitors, customers, market signals, news, hiring activity, financial information, and other sources.
Most business AI tools help answer questions.
We wanted to build something different:
What if AI could investigate whether the assumption behind the question is actually true?
That idea became Business Intent — an AI strategic intelligence system designed to test business assumptions against real-world evidence, identify contradictions, and turn those findings into actionable decisions.
What it does
A user gives Business Intent a company, strategic question, or business assumption.
For example:
“Is expanding into the European market actually our highest-leverage growth opportunity?”
Instead of immediately generating an opinion, Business Intent breaks the question into smaller research problems and investigates the surrounding reality.
It can look for signals such as:
- Market trends
- Competitor activity
- Customer signals
- Company announcements
- Hiring and expansion activity
- Industry developments
- News and public information
- Supporting evidence
- Contradicting evidence
The system then brings these signals together to determine:
What supports the assumption?
What contradicts it?
What might we be missing?
What should we investigate or do next?
The key idea is that contradiction is not treated as noise — it is treated as a signal worth investigating.
How we built it
Business Intent is built as an agentic intelligence pipeline rather than a single LLM prompt.
At a high level:
Strategic Question → Question Decomposition → Research Agents → Live Evidence → Evidence Analysis → Contradiction Detection → Synthesis → Strategic Recommendation
The system separates research from reasoning so that the final answer is grounded in newly gathered evidence rather than relying only on the model's pretrained knowledge.
The intelligence layer is designed around a simple principle:
Don't just generate an answer. Build an evidence-backed model of the situation first.
This allows the system to investigate multiple dimensions of a business problem independently before synthesizing them.
A concrete example
Suppose a company believes:
“Enterprise customers are our best growth opportunity.”
A conventional AI assistant might help the company create an enterprise sales strategy.
Business Intent takes a different approach.
It first asks:
“What evidence would have to be true for this assumption to actually be correct?”
It then investigates the relevant signals and compares evidence supporting the hypothesis against evidence that challenges it.
The result can expose situations where the evidence does not match the company's original belief.
That creates a much more useful output:
Assumption → Evidence → Contradiction → Alternative hypothesis → Recommended action
Why this matters
Businesses already have dashboards showing what happened.
They have analytics tools explaining what happened.
They have AI tools that can summarize information.
But strategic decisions often fail somewhere earlier:
the underlying assumption is wrong.
Business Intent is designed to attack that layer.
Instead of asking:
“How can we execute this strategy better?”
it asks:
“Should we still believe this strategy in the first place?”
That is the problem we are trying to solve.
What we learned
Our biggest lesson was that building an intelligent AI system is not primarily about making an LLM generate better text.
The difficult part is designing the system around the model.
We had to think about:
- How to decompose ambiguous strategic questions.
- How to identify which evidence is actually relevant.
- How to distinguish evidence from assumptions.
- How to handle conflicting information.
- How to prevent confirmation bias.
- How to turn research into an actionable decision.
This led us to think about AI less as a chatbot and more as a system that can:
Observe → Investigate → Challenge → Update → Recommend
Challenges
The biggest technical challenge was that real-world business intelligence is messy.
There is rarely one dataset containing the answer.
Information is distributed across many sources, changes over time, and frequently contradicts itself.
We therefore had to design the system so that it could preserve the original strategic question while allowing multiple research paths to investigate it.
Another challenge was avoiding the failure mode where an AI simply searches for evidence supporting the user's initial hypothesis.
Business Intent is intentionally designed to look for counter-evidence as well as supporting evidence.
What we built
The result is an early working prototype of an AI strategic intelligence system.
A user can provide a strategic question and receive an investigation that connects:
Question → Research → Evidence → Contradictions → Intelligence → Action
Our goal is to evolve this from a research assistant into a continuously updating business reality model that helps companies detect when their understanding of the market is diverging from what is actually happening.
The bigger vision
We believe the next generation of business AI should not simply make existing decision-making faster.
It should make the decision-making model itself better.
Business Intent is our first step toward that vision:
AI that doesn't just help you make a decision — it tries to find out what you might be getting wrong before you make it.
Built With
- ai-agents
- analysis
- data
- docker
- fastapi
- github
- google-cloud
- google-gemini
- javascript
- llm
- natural-language-processing
- python
- react
- restapi
- typescript
- web-scraping
Log in or sign up for Devpost to join the conversation.