Build an AI Agent to Replace 6–24 Hours Per Week of Manual Email Work (Safely) Ep. 95
Hosts:
Justin Shelley - https://www.phoenixitadvisors.com/
Mario Zaki - https://www.mazteck.com/
Bryan Lachapelle - https://www.b4networks.ca/
Joshua Holloway - https://7thdi.com/
A real AI agent watched a bid inbox, parsed attachments, filed everything, and cut 6 to 24 hours a week of manual work. Here is how they kept it from going off the rails.
Justin Shelley - https://www.phoenixitadvisors.com/
Mario Zaki - https://www.mazteck.com/
Bryan Lachapelle - https://www.b4networks.ca/
Joshua Holloway - https://7thdi.com/
A real AI agent watched a bid inbox, parsed attachments, filed everything, and cut 6 to 24 hours a week of manual work. Here is how they kept it from going off the rails.
In this episode, Justin Shelley, Bryan Lachapelle, Mario Zaki, and Joshua Holloway break down what “vibe coding” looks like when it is tied to an actual business bottleneck, not a demo. Josh shares a real client build: an agent that monitors mailboxes, reads emails and attachments, moves files into a consistent folder structure, and extracts key data into Excel as a transitional step toward a dashboard.
They also get candid about the risks. Models change, context breaks, and agents can misinterpret plain language. The group talks about guardrails, narrow task design, approvals, and why you should hard-code what you can so AI only handles the parts that truly need AI.
There is also a practical hiring angle: instead of filling a $95K to $125K role that was open for 6 to 12 months, the company can potentially hire a more junior person who can work with the system, while the business uses the agent to move faster, reduce mistakes, and take on larger opportunities.
What you’ll learn
- How an “email + attachments” agent can save 6 to 24 hours per week by parsing bids, filing documents, and extracting data
- Why consistency wins: how a repeatable folder structure made automation dramatically easier
- How to add a “go or no-go” decision step using business criteria, with a human proofing loop
- Why agents can degrade over time, and how to reduce risk with narrow prompts, hard-coded steps, and guardrails
- A real warning story: how AI can accidentally propose super-admin access, and what to do instead
Subscribe if you want practical, plain-English cybersecurity and AI systems that reduce risk and save real time, not hype.
If you want help designing AI automations with the right security boundaries, or figuring out where AI can remove bottlenecks in your business without creating new risks, Phoenix IT Advisors can help. Schedule a consult at PhoenixITAdvisors.com.
Creators and Guests
Host
Bryan Lachapelle
Hi, I’m Bryan, and I’m the President of B4 Networks. I started working with technology since early childhood, and routinely took apart computers as early as age 13. I received my education in Computer Engineering Technology from Niagara College. Starting B4 Networks was always a dream for me, and this dream became true in 2004. I originally started B4 Networks to service the residential market but found that my true passion was in the commercial and industrial sectors where I could truly utilize my experience as a Network Administrator for a large Toronto based Marine Shipping company. My passion today is to ensure that each and every client receives top of the line services. My first love is for my wonderful family. I also enjoy the outdoors, camping, and helping others. I’m an active Canadian Forces Officer working with the 613 Fonthill Army Cadets as a member of their training staff.
Host
Mario Zaki
During my career, I have advised clients on effective – and cost-effective – approaches to developing infrastructure that fosters productivity and profitability. My work has provided me with a broad-based knowledge of business from the inside, with an expertise in areas that go beyond IT alone, ranging from strategic planning to cloud computing to workflow automation solutions.
