Idea Executor

What Held Up

Most people still treat AI like one smart chat window. That works for rough answers. It breaks down when the work needs multiple perspectives, parallel research, and one clear conclusion at the end.

Idea Executor is the proof that a better pattern already exists:

• split the problem across specialized agents
• let them work in parallel
• merge the result into one usable decision

The startup-idea use case is just the wrapper. The real story is what orchestrated AI workflows can do in practice.

What I Built

Idea Executor fans one prompt out to 7 specialized research agents: market sizing, competitor mapping, user personas, technical architecture, business model, risk analysis, and launch planning. Each agent works in parallel on a narrow slice of the problem. A final report layer pulls the outputs together into one report with a verdict, action plan, and verification checklist.

The key architectural choice is model routing by task shape. When OpenAI is available, the 7 research agents use GPT-4o-mini because those steps are narrow, parallel, and cost-sensitive. The final report step runs on the repo's configured default model, which keeps the implementation simple while preserving one clear place to spend more reasoning budget when needed.

Under the hood, the workflow is simple on purpose: parallel generateText() calls for the research agents, followed by one streamText() call for the final report. The UI streams the final answer instead of putting on agent theater. That keeps the surface clean while still proving the architecture.

Patterns Worth Borrowing

• Fan-out, then synthesize: Any workflow with multiple independent dimensions can use this pattern: hiring scorecards, vendor evaluations, research briefs, product reviews.
• Route models by task shape: Narrow research can stay cheap and parallel. The final report step is the right place to increase quality when the workflow needs it.
• Keep the UI output-first: Users care about the final decision artifact. They do not need a complicated agent graph if the report is clear and trustworthy.

Sample Output Note

The example report shown on the public page is illustrative. It exists to show the shape of the output, not to make a current market claim about that category.