guidesWeb Research

Guide: Web Research

Chief is good at research that requires reading many sources and synthesizing — not just retrieving a snippet from a search engine.

A Simple Lookup

Chief, what's the latest funding round for Anthropic?
Just give me the date and amount.

Chief searches, picks a high-quality source, reads the page, returns the answer. Total time: ~10 seconds.

Multi-Source Synthesis

Chief, I need a snapshot of the AI agent landscape for our
board deck. Look at: Adept, Cognition (Devin), Sierra,
Hyperwrite, and Lindy. For each, give me:
- Company stage / funding
- Core product positioning
- Notable customers
- Pricing (if public)

Format as a table.

Chief visits each company’s site, plus their Crunchbase / Pitchbook public pages, plus relevant press coverage. Returns a single tidy table.

Recurring Monitoring

Chief, every Monday at 8am, scan TechCrunch, The Information,
and Stratechery for any new articles about AI agents or
autonomous AI. Post the top 3 in #strategy with a 2-sentence
summary each.

This is a cron job that runs your research on a schedule.

Watching a Specific Page

Chief, watch this page (https://example.com/competitor-pricing)
once a day. If anything on the page changes, DM me a diff.

Chief snapshots the page, compares on each run, and alerts you on change.

Extracting Structured Data

Chief, scrape the team page at example.com/about. For each
person listed, give me their name, role, and LinkedIn URL.
Output as CSV.

Chief generates a schema, runs the scrape, and uploads a CSV. See Web Browsing for the underlying tool.

Citations

Every research answer Chief gives includes a Sources section at the bottom — a numbered list of URLs Chief actually visited. You can verify or dig deeper.

Limitations

  • Paywalls. Chief won’t bypass paywalls. For NYT / WSJ / niche subscriptions, add credentials to the integration vault and Chief can authenticate.
  • Anti-bot walls. Chief uses a stealth browser fingerprint and handles most public-site protections, but some sites are explicitly hostile to all automated browsing. Chief reports clearly when blocked.
  • Recency. Chief sees the live web. There’s no training-cutoff issue for browsed content. (LLM-generated reasoning over the content is still subject to the model’s general knowledge cutoff.)

Pitfalls

  • “Latest” — be specific about a date window. “Latest as of today” or “in the last 30 days” gets you a more reliable answer than just “latest.”
  • Site lists. When you list specific sources, Chief sticks to them. When you say “search the web,” Chief picks based on its own ranking.
  • Token cost. Heavy research with many sources can cost $0.50–$2 in tokens per run. Plan accordingly for daily research crons.
GuidesWeekly Reports