AI Agents

An AI agent is software that takes a goal, figures out the steps, uses tools to complete those steps, and delivers a result — with minimal human involvement. In construction, this means systems that monitor RFPs overnight, read 200-page specs, score opportunities, and have a ranked shortlist ready before your BD team starts their day.

This isn't theoretical. I build and deploy these systems for contractors and industrial companies today.

AI agents work through multi-step processes autonomously — making decisions, calling APIs, querying databases, and composing outputs. Unlike a chatbot that answers one question at a time, an agent pursues an objective across multiple actions until it's done.

Anthropic breaks agentic systems into two categories:

• Workflows: AI models follow predefined paths — input goes in, output comes out, same way every time
• Agents: AI models decide their own approach — they plan, use tools, evaluate results, and adjust course

How Agents Work in Practice

Here's what actually happens when an agent runs — using the example of an RFP monitoring agent I've built:

The agent runs continuously. When it finds a match, your team gets a notification with the score, the reasoning, and links to the source documents. No one had to search a procurement site — it just showed up.

Levels of Autonomy

1. Human-in-the-loop: The agent does the analysis, a person approves the action. Used for bid decisions, contract approvals, and anything with financial risk.
2. Semi-autonomous: Handles routine work independently but escalates edge cases. Good for daily report compilation, subcontractor performance tracking, and scheduling conflicts.
3. Fully autonomous: Runs end-to-end without intervention. Appropriate for data monitoring, report distribution, and document classification.

Start with human-in-the-loop. Earn trust. Then loosen the reins as the system proves itself.

What I Build Agents For

1. Bid Opportunity Monitoring & Scoring

Agents crawl procurement sites, parse PDF specifications, and score each opportunity against your company's criteria — project type, size, location, bonding requirements, and historical win rate for similar work.

What changes:

• BD teams stop manually scanning procurement sites
• Opportunities get scored consistently using your actual criteria, not gut feel
• The agent catches listings that humans would miss due to keyword mismatches or obscure posting locations

2. Estimate Analysis & Historical Comparison

Agents review draft estimates, compare unit costs against your historical data, and flag line items that are significantly above or below what you've bid on similar projects. They also catch missing scope items by comparing the spec against your estimate breakdown.

What changes:

• Estimators get a second set of eyes on every bid before it goes out
• Pricing outliers get flagged before they become change orders or lost margin
• Historical data actually gets used instead of sitting in old spreadsheets

3. Supply Chain & Subcontractor Intelligence

Agents track material pricing, monitor subcontractor performance across projects, and maintain a living database of vendor reliability — response times, price accuracy, change order frequency, safety record.

What changes:

• Subcontractor selection is based on data, not just who you've worked with recently
• Price trends surface early enough to adjust budgets
• Underperforming subs get identified before they cause problems on the next project

When Agents Make Sense (and When They Don't)

Agents are the right tool when the work involves reading unstructured documents, making judgment calls, and coordinating across multiple data sources. If the task is purely rule-based with structured inputs, a script or traditional software is simpler and cheaper.

The sweet spot for agents in construction:

• Processing RFPs, specs, and contracts (unstructured PDFs)
• Comparing bids against historical data (cross-referencing multiple sources)
• Monitoring markets and competitors (continuous data collection + analysis)
• Generating reports that synthesize information from multiple systems

Getting Started

Pick one problem. Something your team spends hours on every week that involves reading documents, comparing data, or monitoring sources. Build an agent for that. Prove it works. Then expand.

Common starting points:

• RFP monitoring — if your BD team is manually scanning procurement sites
• Estimate review — if you have historical data but nobody's comparing against it
• Daily report analysis — if someone spends an hour every morning compiling updates from field reports
• Contract clause review — if your subs are signing contracts without thorough legal review