How I AI – Episode Summary
Podcast: How I AI
Host: Claire Vo
Guest: Marco Casalaina, VP of Core AI Products, Microsoft
Episode: How Microsoft's AI VP automates everything with Warp | Marco Casalaina
Date: March 23, 2026
Overview
This episode of How I AI features Marco Casalaina, VP of Core AI Products at Microsoft and AI futurist. Marco demonstrates how he uses Warp, an AI-powered terminal, not for its mainstream agentic coding use case, but to streamline a range of practical, often tedious, workflows—both technical and everyday. Throughout the conversation, Marco and Claire discuss the growing potential of micro agents, the abstraction of complex interfaces, and the power of ad hoc automations. The episode is packed with real-life demos, time-saving tips, and insights into the way modern AI tools can drastically reduce friction in routine tasks.
Key Discussion Points & Insights
1. Why Warp? Automating the Mundane (02:44–03:57)
- Marco's gateway into Warp came from Microsoft’s PowerShell team:
“It automates PowerShell really well. And so I tried it, and as soon as I started using it for certain things like managing Azure...man alive, this is a really capable tool.” (02:44) - He highlights that while Warp markets itself as a coding agent, Marco’s primary use is leveraging it as a command line agent for DevOps tasks.
2. The Pain of Traditional DevOps (03:15–06:01)
- Assigning granular roles in Azure typically entails repetitive, time-consuming UI actions:
- “I needed to give [my colleague] Azure AI User and Azure AI Project Manager...If I did it with the web interface...this would have taken me an hour." (03:57)
- Warp lets Marco automate the process via simple prompts, cutting tasks from hours to seconds.
- Warp is just as useful for GCP administration, seamlessly operating with both
azandgcloudCLIs.
3. Abstraction and User Experience (06:01–07:54)
- Claire points out the significance of AI tools accessing CLIs and APIs:
- “You can...abstract away all of that front end stuff and just let a user kind of query the system and get what needs to get done.” (06:01)
- The agentic process eliminates Stack Overflow copy-paste loops, streamlining the workflow directly from prompt to action.
4. Making Agents Smarter: Rules & Documentation (07:54–09:42)
- Marco enhances Warp's capability by connecting to the Microsoft Docs MCP server and defining custom rules:
- “Sometimes...I have no idea what role somebody needs...I can...connect it to the Microsoft Documentation MCP server...it’ll go look it up and that makes it work much better.” (07:54)
- “You can give it these rules and MCP servers that help it along. Of course, that's useful for coding as well, but...super useful for these kinds of things.” (08:48)
- Real value comes from prepping these agents with context-specific rules, enabling smarter, error-free automations.
5. Casual Prompting & Conversational Workflows (09:42–10:56)
- Claire observes Marco’s effectiveness with simple, conversational prompting:
- “This is not the most sophisticated prompting...honestly, for most stuff, just taking the step of writing like two or three steps...are what makes things more effective.” (09:42)
Demo Highlights: Everyday Agentic Automation
1. Scanning & Combining Documents With Warp (10:56–15:34)
Use Case:
- Marco scans his daughter’s practice test, splits it into odd/even pages, and merges them—all automated through prompts in Warp.
Key Moments:
- "[Warp] absolutely...does [activate the scanner]. If I were home...my scanner would wake right up and start scanning." (11:59)
- “I just changed ODD to even over here and it just did that...Then I wanted to put it together, so I said, now put together the odd pages and even pages and just make the math practice test out of it...It wrote some Python to do this.” (13:04)
Efficiency Realized:
- Claire marvels at the time saved:
- "Think about how you would have done it before...Instead you get to sit where we all want to live right now, which is...the terminal in dark mode, and just ask it to do this thing.” (14:06)
- Marco customizes his Warp rules, referencing the use of NAPS2 (a CLI scanner for Windows) for seamless hardware API access. (15:34)
2. Video File Size Fixes & File Manipulation (17:40–20:25)
Use Case:
- After recording a massive 1.7GB video file, Marco uses Warp (plus ffmpeg) to compress it and performs other editing actions (like fixing audio levels).
Notable Quote:
- “I said to Warp...use FFMPEG to like make the sound 300% from 7 seconds to 17 seconds. And it totally does.” (18:32)
- Tasks that would previously involve searching for specialized software are reduced to a single conversational prompt.
Insight:
Claire calls file manipulation "a real underappreciated use case of AI," with Marco agreeing that providing agents access to rich files (plus LLM reasoning) unlocks a new level of capability. (19:19)
The Rise of “Ad Hoc Agents” (20:25–22:51)
- Marco coins the term "ad hoc agent":
- “Effectively, each one of these things that I'm doing...I'm kind of creating a little mini agent, an unnamed agent on the fly to do something for me. And that's becoming a trend.” (20:25)
- Claire advocates for ephemeral workflows:
- “Just toss it, ... don't save this script. ... Next time do it over, do it over again. Maybe save a rule so you’re not rediscovering the steps.” (20:58)
Beyond Tech: Administrative & Knowledge Workflows (22:51–27:17)
1. Workflow Agents in Business Contexts
- Marco demonstrates building agentic workflows with M365 Copilot:
- “When I get an email from Claire Vo requesting a meeting at a certain time, check my calendar, if that time is free, send her a 30-minute meeting invite...” (23:07)
- The agent can extract key details, check availability, and send invites—all automated and triggered by email.
- “The line between consuming an agent and building an agent is blurring. This is an agent that builds an agent.” (22:51)
2. Recurring Checks with ChatGPT
- Marco uses ChatGPT to monitor for new podcast episodes automatically, using scheduled ("cron job") agents.
- “Every day, look to see if there's a new podcast by Claire Vo and notify me if there's a new one. ... Daily at 9am it’s going to check...” (26:06)
Lightning Round: Personal Impact & Prompting Tips (27:17–32:48)
1. Impact on Marco’s Workflow
- “This saves me many minutes a day... I set Warp to running that...while it did that, she and I worked on one of the math problems themselves.” (27:55)
- The only manual task left was flipping the paper—AI doing the rest. (28:32)
- Claire notes this lets you "spend your time differently...on higher leverage, more intellectually stimulating" tasks. (28:42)
2. Teaching the Next Generation
- Marco’s daughter is more of a mainstream user—proficient at using built-in AI features (e.g., Canva’s agent), but not a tinkerer. He believes everyone will learn what benefits them most organically. (29:37)
3. Prompting Techniques & Coping With LLM “Stubbornness”
- Marco on troubleshooting misbehaving agents:
- “I will often be like, you moron. I specifically told you not to check in my ENV file, and you did it anyway. ... There is a rule there that says never check in the ENV file...” (30:53)
- He uses persistent rules in Warp, and, for tools without rules, deploys prepared prompts—hotkeyed via AutoHotKey for fast, repeatable execution. (32:24)
- “You’ve created a library of...little snippets that you know are effective...” — Claire (32:37)
Notable Quotes & Memorable Moments
- "Warp is pretty magical, but you can add to the magic and make it work more smoothly." — Marco (00:00)
- "You can actually abstract away all of that front end stuff and just let a user kind of query the system and get what needs to get done." — Claire (06:01)
- "I'm kind of creating a little mini agent, an unnamed agent on the fly to do something for me. And that's becoming a trend." — Marco (20:25)
- "Just get used to ephemeral stuff, like just toss it...don’t save this script...just come back and do it again...” — Claire (20:58)
- “Efficiency for me is being able to just remotely start the scanner from my AI coding tool.” — Claire (12:19)
Episode Timeline
| Timestamp | Segment/Topic | |------------|------------------------------------------------------------| | 02:44–03:57 | Why Warp for automation: The Azure experience | | 03:57–06:01 | Pain points of cloud admin—how command line saves time | | 06:01–07:54 | Abstraction: AI, APIs, and break from Stack Overflow | | 07:54–09:42 | Enhancing agents with rules, MCPS, and documentation | | 10:56–15:34 | Scanning and merging documents with Warp (live demo) | | 17:40–20:25 | Video file manipulation, ffmpeg, and file intelligence | | 20:25–22:51 | Ad hoc agents and ephemeral workflows | | 22:51–27:17 | Building workflow agents in Copilot and ChatGPT | | 27:17–28:32 | Personal impact and time saved by AI agents | | 29:37–30:30 | AI use with Marco’s daughter: teaching and learning | | 30:53–32:48 | Prompting tactics, fixing stubborn LLMs, and hotkeys |
Final Thoughts
This episode spotlights automation at the ground level—reducing friction in both professional and personal tasks with AI-powered command line agents and workflow automations. Marco’s practical, “ad hoc agent” philosophy shows the future is less about building elaborate, persistent tools, and more about leveraging ephemeral, fit-for-purpose agentic helpers. This is a must-listen for anyone interested in actionable, real-world uses of next-gen AI toolchains.
Find Marco on LinkedIn for more updates, blog posts, and video content.