Posts

Streaming is the next category to consolidate within the modern data stack. IBM announced its intent to acquire Confluent.

The deal values Confluent at $11.1 billion, or 10.0x LTM revenue. Confluent commands more than 40% of the Fortune 500 as customers & has grown into a $1.1 billion revenue business.

The founders of Confluent, including CEO Jay Kreps, created Apache Kafka, a streaming technology built inside LinkedIn. Founded in 2014, Apache Kafka now runs at more than 80% of the Fortune 100. Kafka powers real-time data pipelines & stream processing, updating data systems whenever a new event happens. When a taxi ride is booked, a credit card is swiped, or a user likes a comment, Kafka handles the data flow.

Since I watched software engineers using AI, I’ve become jealous.

I’ve seen the most sophisticated software engineers assign 15 to 20 coding tasks in GitHub to an AI. They play foosball. They grab coffee. They return to evaluate the agent’s work.

The agent tackles the same task three different ways. Sometimes it nails the solution on the first try. Other times it needs more input. That engineer has paralleled her time by 10x to 15x.

While OpenAI signed $1.15 trillion in compute contracts through 2035, DeepSeek trained a frontier model for $6 million. This was 2025’s central question : are we building on bedrock or quicksand?

The top 10 posts of 2025 examined some of these topics :

I remember hosting a dinner of sales leaders to talk about AI. I asked them what will your CRM do for you in the future?

Nearly unanimously came the reply : enable salespeople to spend most of their time with customers.

Listening to Lenny’s Podcast with Jeanne Grosser, COO at Vercel, I discovered how some companies are achieving that milestone by transforming their inbound sales teams in six weeks.

For us, we had 10 SDRs doing this inbound workflow, & now we just have one that is effectively QAing the agent. The other nine, we deployed on outbound.

Private equity firms have emerged as the newest distribution channel for AI startups.

While public companies have decreased from 6,639 in 2000 to 3,550 in 2024, PE-owned companies in the US have grown from 1,950 to 14,300. The rate of growth continues to accelerate.

The AI market today is bacon in a hot skillet. Everything is sizzling, moving, & changing at an incredible pace. We’re all watching it closely.

Market share is fluid because no one yet knows what AI can do & the second we think have grasped it, models improve. The Nvidia chip performance & the launch of Gemini 3 the biggest gain ever in Google model performance suggest no simmering ahead.

As long as the underlying models hurtle towards PhD level performance, people will continue to test. How much better is Gemini 3 at coding? tool calling? writing?

Two revelations this week have shaken the narrative in AI : Nvidia’s earnings & this tweet about Gemini.

The AI industry spent 2025 convinced that pre-training scaling laws had hit a wall. Models weren’t improving just from adding more compute during training.

Then Gemini 3 launched. The model has the same parameter count as Gemini 2.5, one trillion parameters, yet achieved massive performance improvements. It’s the first model to break 1500 Elo on LMArena & beat GPT-5.1 on 19 of 20 benchmarks.

70% of production AI teams use open source models. 72.5% connect agents to databases, not chat interfaces. This is what 375 technical builders actually ship - & it looks nothing like Twitter AI.

70% of teams use open source models in some capacity. 48% describe their strategy as mostly open. 22% commit to only open. Just 11% stay purely proprietary.

Ten months ago, DeepSeek collapsed AI training costs by 90% using distillation - transferring knowledge from larger models to smaller ones at a fraction of the cost.

Distillation works like a tutor training a student : a large model teaches a smaller one.¹ As we’ve shifted from knowledge retrieval to agentic systems, we wondered if there was a parallel technique for tool calling.²

Could a large model teach a smaller one to call the right tools?

By Monday lunch, I had burned through my Claude code credits. I’d been warned ; damn the budget, full prompting ahead.

When that still couldn’t solve the issue, I summoned Opus, the biggest & most expensive model, to solve it.