Most pricing pages change without announcement. A tier disappears, a new add-on shows up, the enterprise plan gets a "contact us" button where a dollar figure used to be. If you check manually, you catch it days or weeks late. If you set up a diff-checker, you drown in noise every time a footer link or cookie banner shifts.
We built Watching Agents at Inithouse to handle exactly this kind of ongoing monitoring. Watching Agents lets you deploy an AI agent to watch any question about the future. It builds hypotheses, tracks evidence in real time, and alerts you when things change.
This post walks through one specific use case: pointing an agent at a competitor's pricing page and getting meaningful alerts instead of raw diffs.
The setup takes one question
You don't configure Watching Agents by writing scraping rules or CSS selectors. You give the agent a question in plain language.
For a pricing page, that question might be: "Will [competitor] change their pricing structure or add a new plan tier in the next 90 days?"
The agent takes that question and does three things:
Builds initial hypotheses. Based on public information it can find (the current pricing page, recent blog posts, press mentions, community threads) it forms a baseline. Something like: "Current structure is three tiers at fixed monthly rates, no usage-based component, last changed ~8 months ago."
Sets up an evidence log. Every time the agent checks, it records what it found. Not a raw HTML diff but a structured note about what's relevant to the question. "Pricing page still shows three tiers. No new blog posts about pricing. Competitor's CEO mentioned 'simplifying our plans' in a podcast published yesterday."
Scores a probability. The agent maintains a running estimate. It starts wherever the evidence points and moves as new signals come in.
What makes this different from a page-change alert
A basic monitoring tool would ping you every time any element on the page changes. Cookie consent text gets updated? Alert. A testimonial rotates? Alert. The actual pricing stays identical, but your inbox fills up.
The agent approach filters through the question lens. It distinguishes between a CSS change that moved a button two pixels to the right and a structural change that removed the free tier. Only the second one matters for the question you asked, so only the second one triggers a notification.
We tested this internally before shipping the feature. We pointed agents at five SaaS pricing pages we track as part of our own competitive research. Over a four-week period, a raw diff monitor would have fired 40+ alerts across those five pages. The agent sent us three. All three were real changes: one company added a startup program, one quietly raised their entry-tier limit from 5 to 10 seats, and one added annual billing that wasn't there before.
The evidence log is the real product
The alert is useful, but the evidence log does more ongoing work. It builds a timeline you can scroll through:
- Week 1: Pricing unchanged. Competitor posted a hiring ad for a "pricing strategist."
- Week 2: Pricing unchanged. Glassdoor review mentions "big pricing overhaul coming in Q4."
- Week 3: Pricing page A/B test detected. Some visitors see a fourth tier labeled "Scale."
- Week 4: Fourth tier now live for all visitors. Free trial extended from 14 to 30 days.
Each entry links to its source. The hiring ad, the Glassdoor review, the cached page variants. All traceable. You're not trusting the agent's summary blindly; you can click through and verify.
This matters because pricing intelligence isn't just about catching the moment of change. It's about understanding the trajectory. A company that hires a pricing strategist, floats trial balloons in job postings, and then A/B tests a new tier is behaving differently from one that drops a price cut with no warning. The evidence log captures that difference.
Avoiding alert fatigue
One failure mode we designed against: the agent that cries wolf. If you deploy 20 agents watching 20 competitors, you need the signal-to-noise ratio to hold up at scale.
Two mechanisms keep it in check:
Deduplication. If the agent already logged a signal (say, the pricing strategist job posting) it won't re-alert you when the same posting appears on a different job board. The evidence is already in the log.
Relevance gating. The agent filters every signal against the original question. A competitor's engineering blog post about database migration? Irrelevant to a pricing question. A competitor's CFO interview mentioning "value-based pricing"? Relevant. The agent makes that call before anything reaches your notifications.
When to deploy this vs. a simpler tool
Not every monitoring job needs an AI agent. If you literally just want to know whether a specific number on a specific page changed, a cron job with a hash check works fine and costs nothing.
The agent approach earns its keep when:
- The question spans multiple sources (pricing page + blog + social + job postings)
- You care about why something changed, not just that it changed
- You need the context around a change to act on it (did they raise prices across the board, or just for new customers?)
- You're watching enough competitors that raw alerts would overwhelm you
We use Watching Agents for our own competitive monitoring across the Inithouse portfolio. When we spot a pricing shift in a space where we operate, the evidence log tells us whether it's a one-off correction or the start of a repositioning, and that changes how we respond.
If you want to try this yourself, you can deploy your first agent at watchingagents.com. The setup is one question and takes about 30 seconds.
Top comments (0)