By Tela Mathias
I love being at events like this, I feel like I have met “my people”. We are this weird, eclectic, smart, funny, and super enthusiastic bunch of nerds. Just really nerdy. And I love it. We are just a medley of misfits. So many great things at Day 2 – but the two major highlights were at the beginning and the end. Simon Willison is a hilariously competent and compelling speaker, and definitely part of our medley of misfits. And closing the day with Greg Brockman was an absolute inspiration. The theme of yesterday was the “the power of optimism”, but maybe that’s just because I’m an optimistic person.
Wish I new his name, but the demo guy at Microsoft was ON POINT. He showed what you can do with Github Copilot and I have to say – wow. The intersection of spaces, Jira, agents, and agent task assignment really told a good story. Imagine that an agent is just another team member, logged in and working like you or me. Imagine that you could assign a task to an agent, readme file generation was the example they used, and then the agent does the work and updates the work item. Now imagine that you need a team member to do a machine learning model, yeah you can assign that to a coworker agent too.
This definitely made we want to make sure that we are making maximum use of the full set of Microsoft capabilities at the team level. It was not, however, enough to make me move from AWS Bedrock to Azure AI Foundary. Maybe I’ll regret this decision at some point but I’m sticking for now. You’re welcome Amazon Web Services.
Talk about overcoming adversity, Sarah Guo is a presentation boss. None of the technology was working, AV was a hot mess, and honestly, they barely figured it out. She used the time with mastery, and I was riveted by her take.
Sarah is the founder of Conviction, an AI venture capital company. She was speaking on the state of startups in 2025 and providing practical advice. One of the things they are very interested in and encouraged is to think about (you know how VC loves their analogies) is: “Cursor for [_X_]”. In our case it would be “Cursor for Compliance” although that’s not where we are yet, but we will be.
One of the reasons Cursor has been so successful is because it was built by engineers for engineers. And engineers know engineers. She out the cherry on top of what we have been hearing for the past six months, content is king. Knowing your customer, knowing your domain space, really building what you know for and market you know – that continues to be the moat.
I loved the idea that building the ironman suit is the bath of least frustration. Start with what you know, you can always make it better. Sarah Gua is a BOSS. Loved her.
I had, sadly, never heard of Simon Willison. He was falling out of your chair funny. I love his personal eval, “product an SVG of a pelican riding a bike”. This reminded me of Ethan Mollick and his “otter taking a plane ride” eval. So Simon was there to talk about the past year in LLMs, but there was too much so he skinnied his scope down to the past six months.
The reason he uses the pelican riding the bike is because (a) he’s tired of the other benchmarks and has lost trust and (b) it’s a great test because it requires technical prowess in producing the SVG, the pelican has very difficult anatomical structures that are incompatible with riding a bicycle, and the bicycle seems simple but is actually a challenge for humans to illustrate due to its interesting geometry.
Some of the key points made clear in the past six months:
This one was near and dear to my consulting roots. I had never heard of the company, but I really resonated with what they were talking about. They talk about the staffing models – traditional pyramid v. inverted pyramid (relying on junior staff to do most of the work v. relying on senior staff to do most of the work). And their hypothesis on the future for professional services is the inverted pyramid in the center, with traditional pyramids of agents at each side. This makes a lot of sense to me. Not sure I would have illustrated it this way but intuitively, it’s the right move.
I was surprised that they did not discuss voice agents more specifically, I think the opportunity there is massive. Imagine if you would interview and entire company in, like, two hours. Yeah, voice agents. I’m here for it.
I’m so glad that we pivoted away from automating software development because man, Windsurf has pretty much crushed that. This one was personal, when we started this AI journey in December of 2023, I was hell bent on “push button, get software”. Many of our early research meetings were about this idea of automating software development. As we learned more, and really listened to our industry feedback, we realized we needed to be a lot more specific and much closer to the market – hence mortgage compliance change management, of which software development is a key part.
And that was a really good pivot. Windsurf is going to absolutely crush this space. Their vision is ridiculously bold – to be everywhere, doing everything, all at once. And I believe them when they talk about how they intend to do it. I think they will crush everyone, they certainly would have crushed my original product concept. Phew – dodged a bullet there.
Greg really gave Jenson a run for his money on being my idol and personal hero (#jensenisstillthegoat). I’m slightly embarrassed to admit I did not know him before this closing keynote. Well, I certainly won’t forget him. He was absolutely inspiring. This will be the subject of a separate article. Too short on time to do it justice.
