With another Summit in the books, I'm reflecting on the week while I make the long journey home.
There's a lot to unpack from the conference, but I think I've already summed the week up quite nicely in the Summit Discord server with the following:
AI AI AI AI AI AI AI
YOURE NOT DOING ENOUGH AI
LOOK AT ALL THE AI THINGS EVERYONE IS DOING
DAMN GOOD GUI GUY
Graph API and DSC have received the following incremental upgrades you should know about.
AIAIAIAIAIAIAIAIAIAIAIAI
🤖 🤖 🤖 🤖 🤖 🤖 🤖
Care to guess what the topic of the week was?
Of course it's no surprise that one of the major topics that found its way weaving in and out of all the traditional talk subjects - when it wasnt busy being the main topic itself - was AI.
Two things stuck out the most to me:
• The size if the gap between the people who are really putting AIs pedal to the metal and the ones who are still staring wide-eyed into the AI abyss(🙋🏻♂️)
• The size of the gap between what the tools and pipelines were capable if last year at Summit to this year at Summit
Those two things have a scary force multiplier to them when you think about it.
Said another way: I was behind the pack last year, and the lead runner has gained distance on me and has since wrote AI agents that run for them.
Are you feeling left behind? Scared its too late?
I know I am.
....and yet.
I lied to you, because theres a third thing I noticed, one I noticed within myself, at Summit this year:
• I haven't felt so f#$/!?@ inspired and fired-up in years
Not exactly the conclusion you'd expext me to draw, and I have to say I ageee.
But, how could I not be inspired?
I just spent a week with some of the smartest kids at the playground and they are doing things that just never existed until now. They're turning either themselves as individuals or their small teams into rocket fueled creators and problem solvers.
They're building tools with skill sets they didn't possess in recent memory, that they learn at an accelerated rate with AI or that they didn't need to become an full expert in to be able to assess the generated code and iteratively tease out more functionality and increased stability.
They aren't even building the slide decks themselves any more and they are some of the most engaging and well laid out decks I've ever seen!
Seriously, they have nice decks.
Lets talk data. Where is this inspirational sunshine coming from and how is it dispelling the creeping shadowy doubt of being "left behind"
Comforting words from Jeffrey Snover himself. He let us all know that were actually insane, and some how that made it feel like we were going to be okay.
The basic premise: Insanity means your world-view doesn't align with reality.
Our world-view hasn't really changed, but reality has. AI tools, properly used and structured, are making it so all those things you once thought you couldn't do are actually very much in reach.
Jeffery thought he was a CLI guy. He used intelligence, magic and determination to bring PowerShell into the world and with it he changed the lives - or at least careers - of so many people. So when he needed a GUI for his small retirement hobby of investigating the toughest technological questions around AI as a Fellow at Harvard Law he naturally thought I'm not a GUI guy, I can't build a GUI.
But Jeffrey Snover, **was insane. **
Reality quietly shifted the overton window of what is possible and what isn't, and when he woke up out of his insanity after vibe-coding a GUI for his project in five weeks which may very well have taken a team five months to build he found his sanity, and his world view finally aligned with reality.
He is a GUI guy
Not only that.
He is a "Damn Good GUI Guy"™️
So maybe im not behind, maybe I'm just insane. I must be, since that potential insanity appears to be giving me hope.
Yeah you read that right settle down its a very respectable and incredibly useful security auditing tool.
(We can all name things whatever we want now. We're insane, remember?).
You'll definitely want to visit https://GloryRole.com (please type it carefully) and check it out.
I find this project to be so insanely inspiring. The use of esoteric data science concepts combined with audit logs creates alchemy used to distill down the roles and permissions applied within your organization not only to least privilege, but also to a mathematically sound guess at the least number of roles you need to accomplish it, has got to be fueled by exactly the reasons why we do any of this. Its not just doing the job and getting the outcome, its loving the craft and asking what would we do if we could do anything to solve this problem.
I don't want to make assumptions about Frank, Danny's Regular Intelligence coding chops, but I'm pretty sure learning the data science and implementing it and not having it take the majority of their careers thus far to do so probably had a little bit to do with their mastery of the AI pipeline.
And if we trust in reality, instead if our own insanity, just maybe we could do something like that too.
Did I mention they have great decks?
Its becoming increasingly clear where my knowledge gaps have been in AI, and the more I learn the more I can start to build my AI tooling pipelines.
Gilbert Sanchez was explaining a slice of what he's got going on in his pipeline. He's got a lot of projects out there in the world, and the age old issue for someone who is building things for the community is the correlation between the amount of time spent maintaining the things you've made, and the amount of time you have to make new things.
Thats why it was so intriguing to hear of how he's using AI in his maintenance of PSSAKE, a build automation tool for your PowerShell development.
He's put in the work to develop style guidelines and skills that AI can use for everything from bug fixes and feature additions to changelog entries and full release pipelines.
It's the kind of thing about AI tool sets that is made so much more impressive when you watch them ORCHESTRATE IT ALL FROM THEIR PHONE.
THEIR PHONE.
Gilbert is doing all of this from his phone in front of me as we chat.
Its not altogether crazy that someone was able to use the supercomputer they carry in their pocket to accomplish some amazing things, but in a world where this branch of technology feels so much more like astonishing feats of wizardry and magic, there's something about doing it with your phone that is surprisingly effective at driving home the magnitude of the change in reality that AI brings to the table.
Last year felt like there was a very obvious elephant in the room. If AI is this powerful and capable of developing its own solutions on the fly then what are we to make of the very skill set for designing and creating tools that we are at this conference to discuss? Are we simply a group of experts at something that no longer matters?
Turns out there's actually a lot of hope in that department. During the state of the Shell discussion Steven Bucher and Michael Greene from Microsoft talked at length about the role PowerShell can play in AI.
The interesting thing about how AI has developed the skill set to be so useful and capable is that it was trained on data that was supposed to be training us.
Documentation.
Blogs.
How-To Guides.
Written language.
This gives us two distinct advantages in this new technological mindset shift.
The first is one of the pillars that serve as the very backbone of what makes PowerShell an accessible tool: Verbosity.
Its always been standard practice to adopt a Verb-Noun naming structure when creating CMDlets, and to make the function describe exactly what its going to accomplish, even if it makes for a command that requires a few days of your time to finish typing. Its exactly that natural language verbosity that lends itself so well to the exact learning style that AI was built on.
The same can be said of one of the most important elements of a good CMDlet which is comments.
Specifically, comment-based help.
The organized and structured natural language comment-based help being embedded in the command along with detailed explanations of each parameter and examples that include multiple combinations of parameters mean that AI can easily get a firm grasp on even custom functions it wasn't trained on (aka, your functions!).
The second pillar sounds at first like it would be in direct conflict with the first: lower token usage.
Cloud based AI is billed by tokens, which is almost like paying by the syllable, so you might be tempted to naturally conclude that having AI use and produce PowerShell would lead to a higher cost than interacting directly with available APIs. Steve and Michael showed us some examples that proved that's not always the case.
Commandline input and output often ends up being more succinct than having to repeat long URIs over and over for each call.
It also gets to benefit from something that Michael called "Hero Functions", or commands that eschew the usual advice of 'a function should do only one thing and do it well', in favour of smashing four or five things into one command that serves a single over all use-case. The example given was New-AzVM, which pieces together the work of multiple CMDlets to create networks, resource groups, storage, network rules etc. that are all needed to get a VM up and running in Azure. This would normally run counter to "the right" way of creating commands but ends up being a major factor in keeping down the costs of the answers AI is returning.
Overall that means that we aren't learning a dying art. It's not a waste to and keep our skills sharp through practice and through events like Summit but rather it turns out AI is just another customer that needs us to build tools and review its use of them.
So this year, I'm choosing:
Inspiration over Fear.
Acceptance over Denial.
Sanity over Insanity.
It's not over for us, we haven't reached the Summit.
We have a Glorious Role yet to play and a lot to learn so we can all continue to be PowerShell Heroes.
✌🏻