Five important Runtime stories from 2025

Today: We run out the clock on 2025 with a look back at the year in enterprise AI, the latest enterprise moves, and the last Runtime roundup of the year.

Five important Runtime stories from 2025
Photo by David Becker / Unsplash

Welcome to Runtime! Today: We run out the clock on 2025 with a look back at the year in enterprise AI, the latest enterprise moves, and the last Runtime roundup of the year.

(Please forward this email to a friend or colleague! And if it was forwarded to you, sign up here to get Runtime each week.)


The way we were

The traditional way of filling the tech news vacuum that arrives every year the week before the holiday break is to offer a guess at what might happen next year, partly because it can be fun and partly because tech people prefer looking ahead over reflecting on the past. But there's a reason venture capitalists spread their bets on the future far and wide; nobody really has any idea what's going to happen next.

Instead, let's revisit five key stories from the past year that explored how enterprises grappled with the problem of actually getting the generative AI tools and agents that promise a new era of business productivity to work. Here's one prediction for 2026: While things are getting better, the struggle will continue.

Up to code: If there was one generative AI or agentic tool that found traction in 2025, it was the AI-powered coding assistant. Cursor quickly established itself at the vanguard of this movement while Google snapped up much of the team behind its rival Windsurf and AWS introduced Kiro, which felt like it was mentioned as many times as any single other product during re:Invent 2025.

But it was Microsoft, GitHub, and its longtime flagship coding tool Visual Studio Code that paved the way for the agentic coding explosion, and as the year ends it's not clear how much longer that first-mover advantage will last. We previewed that outcome in January; "Any time when you have a shift, there will be winners who emerge," said Igor Ostrovsky, co-founder of Augment Code, which is also building an AI coding assistant.

Format follows function: After beginning to wrap their heads around the concept of AI agents in late 2024, companies quickly realized this year that they weren't going to make much progress with agents until they got their corporate data organized and modernized. Databricks was in a great position to capitalize on that shift, ending the year on pace to record nearly $5 billion a year in revenue and worth $134 billion.

But after buying a pre-revenue startup called Tabular in 2024 for $2 billion in hopes of bridging the compatibility gap between Databricks' Delta Lake open-source table format for corporate data and the industry's preference, Apache Iceberg, Databricks CEO Ali Ghodsi conceded in May that customers might be less interested in formats and more interested in outcomes. "This sort of war on formats — do you put the bit this way, or do you store the invariant this way — maybe it's not needed. Maybe all that's needed is you talk to the catalog, you tell it, 'I want this data', and then it gives it back to you in the format you want it," he told Runtime.

AI protec but it also attac?: One of the first stories in the history of Runtime was a look at how generative AI was impacting the way security professionals approached their jobs, as pessimists braced for a new onslaught of attacks while optimists jumped at the chance to infuse the technology into their defenses. Two years later, according to Amazon Chief Security Officer Steven Schmidt, the defenders are still way ahead of the attackers when it comes to using generative AI effectively.

Schmidt made those comments in the middle of the year, but they stood up over the second half of the year, AWS CISO Amy Herzog told Runtime at re:Invent 2025. While Anthropic turned a lot of heads in November by revealing how its Claude models were used to launch several dozen attacks, security pros quickly dismissed that revelation as a publicity stunt that didn't really show how attackers could do much beyond the capabilities of what existing tools and techniques allow.

Design thinking: After decades of building and using applications designed around windows, menus, and pull-down, generative AI offers a chance to rethink how we interact with software to accomplish a task. Can the applications of the future get away with just a prompt box and a results page?

Probably not, according to experts interviewed on this topic in August, but if enterprise generative AI apps stick they are likely to look very different compared to today's user experiences. "In the future, there won't be one standard way everybody experiences a certain software," said Amol Ajgaonkar, CTO at Insight.

R U up?: The topic came up again and again throughout 2025 during conversations about building enterprise generative AI apps; the APIs that developers rely upon to tap into models from OpenAI and Anthropic are really flaky. "It is what it is," Zencoder co-founder and CEO Andrew Filev told Runtime. "People are ready to live with it today because, one, they have no choice, and second, because of the productivity benefit" they get when everything works as it should, he said.

Both OpenAI and Anthropic are aware of their reliability problems, but solving them is easier said than done thanks to surging demand for their services and the fact that deploying AI models is much more complex compared to serving traditional workloads from traditional hardware. As the year ends, it's not clear that either company has made much progress.


Looking for a way to support independent tech journalism this holiday season? For $10 a month, you'll help us continue our mission to bring reliable and actionable coverage of this vital sector of the economy and gain access to supporter-only features currently in the works, such as an exclusive discussion and events forum.


Enterprise moves

Peter DeSantis is the new … something at Amazon overseeing the company's AI model development, custom silicon work, and quantum computing efforts, replacing Rohit Prasad, who is leaving the company.

Bjorn Hovland is the new president of CIQ, a promotion from his previous role as chief operating officer of the company behind Rocky Linux.

Ori Yitzhaki is the new chief product officer at Sonar, joining the software QA company after product leadership roles at AppsFlyer and JFrog.

Ronald Richardson is the new chief revenue officer at Leaseweb, joining the Dutch cloud infrastructure company after sales leadership roles at Banyan and Microsoft.


The Runtime roundup

Google and Meta are working together on a new project to make it easier to run PyTorch on Google's TPUs, according to Reuters, which could weaken one of Nvidia's longstanding advantages in the AI accelerator market.

Anthropic released its Agent Skills format, designed to give agents more context to be successful, as an open standard with buy-in from Microsoft, Atlassian, Stripe, and others.

Docker released its Docker Hardened Images as an open-source project, setting up a showdown with Chainguard as another source of vetted software images for developers to use in their own projects.

Snowflake customers endured a widespread, hours-long outage this week after releasing a faulty update to its flagship product, according to The Register.


Thanks for reading — Runtime is off for the holidays and the rest of 2025, see you Tuesday January 6th!

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to Runtime.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.