Building Tendos AI: How an Agent Swarm Turns Construction Emails into Quotes

This episode of Just Now Possible features Tendos AI's leaders discussing how they automate the tendering workflow for construction manufacturers, where teams manually parse huge bid-request PDFs, match products, price them, and draft quotes under tight deadlines. The conversation traces their path from a narrow radiator-matching prototype for one design partner to a multi-agent system that handles email categorization through offer generation. It matters for product teams because it shows how to validate an AI opportunity through domain expertise and on-site observation, and how to build trustworthy agentic products with per-agent evaluation, review agents, and human feedback loops. The source text provided is only the show notes and chapter outline; the full transcript is paywalled, so this metadata reflects those notes.

01Key takeaways

• Start with one narrow use case and one design partner to prove value before expanding scope.
• Spend time on-site watching users work to uncover the real workflow before designing automation.
• Build your own web interface when it gives control over UX and a route to full automation.
• Evaluate each agent separately so you can see exactly where performance changes and debug quickly.
• Add a review agent that checks other agents' outputs, similar to code review, before work reaches humans.
• Treat customer requests to replace existing tools as strong evidence of product-market fit.

02Key sections

The tendering chain problem

Construction manufacturers must manually open bid emails, parse massive PDFs, select relevant products, look up prices, and draft quotes before deadlines. The process is tedious and error-prone, making it a ripe automation target.

Starting narrow and validating

Tendos began with radiator requests for a single design partner, using the CEO's construction background to identify and validate the opportunity. The narrow scope proved value before expansion.

Owning the interface

Building a web application rather than integrating into legacy systems gave the team control over UX and a path toward full automation.

Multi-agent architecture and evaluation

Specialized agents collaborate, including a review agent that checks other agents' work before human review. Each agent is evaluated independently, and custom observability tools were built when off-the-shelf options fell short.

Customer pull and the path to self-learning

Customers asked Tendos to replace their CPQ software, a strong product-market fit signal. Human-in-the-loop feedback is pushing the system toward self-learning.

03From the post

“Listen to this epsiode on: Spotify | Apple Podcasts When a construction company receives a bid request, someone has to open that email, parse the attached PDF (sometimes 1,800 pages describing an entire building), figure out which products are relevant, look up pricing, and draft a quote—all before the”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.