The business context files are what make this click. Most 'AI research agent' tutorials give you a generic tool. Loading your own niche, audience, and competitor watchlist before every session is what turns it into an agent that actually knows what it's looking for. I built something similar pointing at program and stakeholder context instead of content, same architecture, completely different output quality once the context files are right.
The 27-tab research spiral is painfully familiar — I've lost entire afternoons to that exact copy-paste gymnastics between browser, Docs, and Notion before forgetting what I even found.
What sold me here is piping Perplexity searches and Firecrawl scrapes into structured JSON that feeds directly into a content pipeline. That's the leap most "AI research" tutorials skip - they stop at "get answers" instead of building the automated loop. I went down a similar rabbit hole building my own Claude Code agent that handles research, drafts, and deploys autonomously: https://thoughts.jock.pl/p/wiz-personal-ai-agent-claude-code-2026
Curious about one thing - how do you handle the agent drifting off-topic when Perplexity returns tangentially related results? Any guardrails built into the prompt, or do you filter in the JSON schema?
The business context files are what make this click. Most 'AI research agent' tutorials give you a generic tool. Loading your own niche, audience, and competitor watchlist before every session is what turns it into an agent that actually knows what it's looking for. I built something similar pointing at program and stakeholder context instead of content, same architecture, completely different output quality once the context files are right.
Yes, the idea is the same architecture can be extended into different use cases, much like what you did. Amazing!
The 27-tab research spiral is painfully familiar — I've lost entire afternoons to that exact copy-paste gymnastics between browser, Docs, and Notion before forgetting what I even found.
What sold me here is piping Perplexity searches and Firecrawl scrapes into structured JSON that feeds directly into a content pipeline. That's the leap most "AI research" tutorials skip - they stop at "get answers" instead of building the automated loop. I went down a similar rabbit hole building my own Claude Code agent that handles research, drafts, and deploys autonomously: https://thoughts.jock.pl/p/wiz-personal-ai-agent-claude-code-2026
Curious about one thing - how do you handle the agent drifting off-topic when Perplexity returns tangentially related results? Any guardrails built into the prompt, or do you filter in the JSON schema?
Thank you @Jessica Drapluk for sharing, means a lot!