General Contractor

Back to discussions
Expand all | Collapse all

How I'm Actually Using AI Agents in Real Workflows (and Where They Break)

  • 1.  How I'm Actually Using AI Agents in Real Workflows (and Where They Break)

    Posted 2 days ago

    I've been spending a lot of time recently experimenting with AI agents in real workflows (not just demos), and I wanted to share what I'm seeing and also learn from others here.

    Specifically, I've been using tools like Claude Coworker, Nemoclaw (still in alpha), and Claude Code to handle things like:

    • navigating systems

    • pulling data from various systems e.g. Slack, Hubspot, Email, Calendar and asking the agent to take draft action

    • stitching together multi-step processes

    • Reconciling data across software (e.g. Procore vs Spectrum)
    • Writing code

    Some of it is surprisingly useful today. Some of it breaks pretty quickly once things get messy.

    One area I've been thinking a lot about - especially for construction finance - is security and data privacy.

    These agents are:

    • accessing multiple systems

    • handling potentially sensitive financial/job data

    • sometimes making decisions or taking actions across tools

    So questions start to come up like:

    • Where is the data actually going?

    • What gets stored vs. transient?

    • How do you control access and permissions across systems?

    • What's acceptable vs. risky in a finance/accounting context?

    I don't think there are fully clear answers yet, especially as these tools move from "assistant" to more "agent-like" behavior.

    One thing that helped me frame this better is a recent research benchmark called GAIA2, which looks at how AI agents perform in dynamic, real-world environments (where timing, uncertainty, and coordination matter). It highlights tradeoffs between reasoning ability, speed, and robustness - which feels very relevant to how these systems behave in practice.

    I'm putting together a small session / roundtable to walk through:

    • what's actually working today with AI agents

    • where they fall apart in real business workflows

    • how to think about security / privacy tradeoffs

    • how this might apply (or not apply) to construction finance teams

    • and a simple breakdown of GAIA2 in practical terms

    This is not a sales webinar - just trying to have a grounded discussion and compare notes with others who are curious or experimenting.

    If this is interesting to you, you can sign up here:
    πŸ‘‰ https://beiinghuman.com/ai-agents-in-construction-finance/

    Would also be curious:
    How are you thinking about security and data exposure when it comes to AI tools in your workflows?



    ------------------------------
    Rishi Srivastava
    CEO
    Beiing Human
    (608) 772-3912
    United States
    ------------------------------


Related Content