Autonomous AI Agent Development

Description

This is a research-adjacent build and we want to be honest about that up front: we are not certain the autonomous version is better than the scripted one, and part of the engagement is finding out. The domain is competitor price monitoring across about 60 retailer sites. Our current scripted scrapers break constantly because the sites change, and maintaining them is a full-time job we do not want to keep funding. The hypothesis is that an agent that can look at a changed page and work out where the price moved to is cheaper to run than a scraper we rewrite every fortnight. Deliver it as an experiment with a measured answer. If the honest conclusion is that the scripted approach wins on cost and reliability, say so and we will pay for that answer.

• Build the autonomous extraction agent with a hard per-run cost ceiling and a stop condition
• Run it head-to-head against the existing scrapers on the same 60 sites for an agreed period
• Measure accuracy, cost per extraction and maintenance hours for both approaches
• Sandbox execution so a misbehaving run cannot touch anything outside its own workspace
• Report the honest conclusion, including a negative one

• Autonomous extraction agent with cost ceilings and sandboxing
• Head-to-head comparison data against the current scrapers
• Cost and reliability analysis with a clear recommendation
• Either a production rollout plan or a written case for not proceeding