How a Chief Engineering Officer Uses AI to Automate His Day

You know the feeling, a customer call is in an hour, and they want deep security details. In today’s issue, we show exactly how to turn a wiki page into a crisp two-slide deck, then how to stand up an event-triggered assistant that does the research for support tickets, so you only do the final 2 percent.

This New Way is a weekly playbook for leaders shipping real AI workflows that move the needle.

👉 Listen to Ali's episode

🎬 Subscribe to our YouTube Channel

These tutorials mirror the live demo from the episode, tailored so any team can reproduce the results with tools like ChatGPT or Claude plus your existing wiki and ticketing stack.

Tutorial 1: Build a two-slide customer deck from your wiki in 60 minutes

Busy execs need speed and quality. This workflow turns a Confluence page into a value prop slide and a sequence diagram that satisfies architects.

1. Connect your wiki to your model

• In ChatGPT or Claude, enable the Confluence connector or MCP server. Grant read access to the specific space or page.

2. Paste a structured system prompt

• Example: “You are a technical product marketer. You write punchy messaging. Read the attached Confluence page on . Create two slides:
  Slide 1: value proposition bullets for buyers.
  Slide 2: sequence diagram of the authentication flow in Mermaid format.”

3. Ask for exact outputs

• Request: slide titles, 4 to 6 bullets, a single-sentence takeaway, and a Mermaid sequence code block named auth_flow.

4. Validate with a quick spot check

• Scan for claims that need citations, versions, or limits. Ask the model to add inline footnotes to any performance or compliance statements and to include source anchors back to the wiki.

5. Move into your deck

• In PowerPoint, use Copilot’s “Insert slide” or copy bullets directly. Paste the Mermaid into a diagram tool that supports Mermaid, export as PNG, then drop it into Slide 2.

6. Tighten for the audience

• Ask the model: “Rewrite Slide 1 for a CTO audience. Keep 45 words total. Replace adjectives with proof points.”

• Then: “Rewrite Slide 2 speaker notes for an architect, 80 words, precise nouns, no fluff.”

Pro tips

• Model choice: use GPT 5 or Claude 4.5 Sonnet for accuracy on technical docs.

• Guardrails: ask for a “hallucination check” that flags any content not found in the source page.

• Keep a reusable prompt template called “Two-slide briefing” inside your workspace.

Watch the deck-from-wiki demo

Tutorial 2: Auto-triage a “missing order” ticket with event-triggered agents

Turn an inbound Jira ticket into a researched draft response. Your assistant looks up the customer, pulls recent orders, suggests the likely issue, and posts a comment for your human approval.

1. Define the trigger

• Create a Jira automation that fires on new tickets with keywords like “missing order.”

2. Route to an workflow tool

• Send the ticket payload to your assistant’s workflow tool (Zapier, Gumloop, Lindy or n8n) with fields: reporter name, email if present, summary, description, and ticket URL.

3. Fan out to domain agents

• CRM agent: resolve the customer record from name or email, return account ID and tier.

• Order agent: fetch last 5 orders, status, shipment IDs.

• Knowledge agent: pull the “where is my order” SOP.

4. Synthesize the case

• Prompt the orchestrator to produce:
  a) the most likely missing order with evidence,
  b) two alternative hypotheses,
  c) a Jira comment summarizing findings,
  d) a ready-to-send customer reply, and
  e) next actions with deep links.

5. Write back to Jira

• Post the synthesized comment to the ticket thread. Leave the customer-facing draft as a separate suggested reply for human review.

6. Add safety checks and evals

• Require human approval before outbound messages.

• Create two or three eval cases, for example “ambiguous identity,” “multiple orders same day,” and “no CRM match,” then run them on every model update.

Pro tips

• Start narrow. Support one ticket type first, then expand coverage.

• Add a confidence score. If below threshold, ask the agent to request one clarifying detail from the customer.

Watch now: See the event-triggered assistant walkthrough

AI tools mentioned

• ChatGPT or Claude (connected to Confluence) for retrieval and drafting.
  

• Microsoft Copilot for PowerPoint slide insertion (alternatively use Gamma!).
  

• Mermaid for sequence and simple charts.
  

• Jira for triggers and ticket comments.
  

• Optional: an orchestrator like Solace Agent Mesh or your preferred agent framework, plus CRM and order system connectors.

Why this matters

Leaders win when quality goes up while prep time goes down. These two workflows create that leverage, and they compound as templates across your org.

👉 Subscribe to This New Way: YouTube Channel

 📩 Was this email forwarded to you? Subscribe here

Until next time,

Aydin Mirzaee
CEO at Fellow.ai & Host of This New Way