Agentic AI makes old worries about lock-in look quaint

Welcome to Runtime! Today: Like its competitors, Google Cloud thinks there can only be one agentic AI management platform to rule them all, OpenAI doubles the price of accessing its latest model, and the latest enterprise moves.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime each week, and if you value independent enterprise tech journalism, click the button below and become a Runtime supporter today.

Become a Runtime Supporter

Choose your fighter

LAS VEGAS — After nearly two years of hype surrounding AI agents and their potential to forever change the way enterprise tech organizations design and build their internal software, things are starting to happen in 2026. But buyers interested in rolling out agents across their companies have a big decision to make that will be hard to undo: Which company will they trust to manage that activity?

Google Cloud executives spent a week in the desert at Google Cloud Next 2026 arguing that it has the best soup-to-nuts combination of enterprise AI technology, from its TPUs to its Gemini models to the new Gemini Enterprise Agent Platform. Putting aside the merits of those claims, enterprise AI appears to be following a similar playbook to the generations of enterprise software advances that came before it; once you picked a platform for your applications and data, you were more or less stuck there.

Gemini Enterprise Agent Platform is the new name for Vertex AI, and "this is our blueprint for the agentic enterprise," CEO Thomas Kurian said in his opening keynote on Tuesday. "It expands on the previous capabilities of Vertex AI and brings new capabilities to enable your teams to build, scale, govern and optimize agents with the same architectural rigor you apply to your most mission critical systems."
There are several components to Agent Platform, including a new version of Agent Development Kit that allows customers to organize agents into "sub-agents" to chain tasks together, as well as Agent Runtime (!), which speeds up the process of launching new agents and lets customers create agentic workflows that can run for multiple days.
Agent Identity specifies a way for customers to attach a "cryptographic ID" to each agent for auditing purposes, which is a primary concern for CIOs and CISO worried about agents going rogue.

Those are all welcome features for enterprises that need a little help getting agents up and running in their environments, but they are far from unique; both AWS and Microsoft released updates to similar tools for building and managing agents on Tuesday. There's a reason why the hyperscalers and enterprise SaaS stalwarts like Salesforce and ServiceNow are all trying so hard to get customers to choose their management layer as the "single source of truth" for AI agents; data has gravity.

"There has to be one agentic platform," said Karthik Narain, chief product and business officer for Google Cloud, in an interview with Runtime.
At one point a few years ago companies thought they could stitch together agent platforms with different models and different providers, but that's too much heavy lifting for most customers, Narain said.
In conversations with potential customers, he argues that even if they want to go with a different provider, "the choice of not selecting Google as that platform is a less expensive choice compared to having multiple platforms to manage your agents, because the agent lifecycle needs to be managed from its identity to safety, security, observability, guardrails, evals, and the reinforcement loop." he said.

However, that means that once enterprise buyers select a platform to manage their agents they're likely to be stuck there for quite some time. And they're being forced to make that decision at a time when no one knows exactly where these tools and platforms will be in a year, and when Anthropic and OpenAI are making a big push to convince companies to work directly with them.

"Instead of trying to build agents that all run one-off, organizations need to create an agent platform where they are able to build, scale, deploy, operationalize, and also have a gallery view of what agents are available for them," Narain said.
At the moment, he's probably right, but it's a little funny that Google Cloud rose to prominence by making the exact opposite argument; that running all your workloads in a single cloud platform was too risky for the modern enterprise.

No small potatoes

OpenAI dropped GPT-5.5 Thursday, the model that was referenced by its code name, "Spud," in new chief revenue officer Denise Dresser's enterprise-focused memo to employees last week. It's the company's most powerful model released to date, and it said in a blog post that "the gains are especially strong in agentic coding, computer use, knowledge work, and early scientific research — areas where progress depends on reasoning across context and taking action over time."

Benchmarking results provided by OpenAI compared GPT-5.5 quite favorably to the state-of-the-art models released so far by Anthropic and Google, although those stats never would have seen the light of day otherwise. However, there's a big catch: "OpenAI has effectively doubled the entry price for its flagship model compared to the previous generation, and again doubled it from there for the most-cutting edge variant of the model, GPT-5.5 Pro," according to VentureBeat.

OpenAI said the model was "more efficient in how it works through problems, often reaching higher-quality outputs with fewer tokens and fewer retries," but the tokenmaxxers of the world will likely just throw more work at it. For the rest of the world, with actual budgets to follow, GPT-5.5 could mark an end of a wild period of AI experimentation and the start of a more cost-focused, disciplined one.

Enterprise moves

Dan Shapero is the new CEO of Linkedin, replacing Ryan Roslansky, who will remain executive vice president of LinkedIn and Microsoft Office for its parent company.

Michael DeCesare is the new president of Oasis Security, joining the agentic identity company after leadership roles at Abnormal AI, Forescout Technologies, and McAfee.

Joe Beda is the new chief technology officer at Stacklock, rejoining fellow Heptio co-founder and Kubernetes author Craig McLuckie at the agentic platform company.

Tim Yarbrough is the new chief financial officer at Checkr, joining the background-checking company after finance leadership roles at ZipRecruiter and Converto.

The Runtime roundup

ServiceNow stock fell nearly 18% Thursday after it reported earnings Wednesday that beat expectations and raised guidance, but acknowledged that the war in Iran disrupted several deals.

IBM was off more than 8% Thursday after reporting a similar earnings story Wednesday, a sign that the SaaSpocalypse is not over quite yet.

Anthropic acknowledged that Claude Code's performance has been off lately due to "three separate changes" that inadvertently caused problems, and said they have been addressed.

Vercel reported that some customer data was compromised in a separate incident prior to last weekend's hack, which is not a great look.

Thanks for reading — see you Saturday!