Skip to content

Runtime: Why building long-haul agents is so tricky

+ Anthropic's new national security advisor, and Deutsche Telekom thinks AI will save it €2.5 billion.

Runtime: Why building long-haul agents is so tricky
Photo by Fabio Sasso / Unsplash

Welcome to Runtime! Today: Restate thinks software developers should spend less time building complicated backends, why Anthropic hired a former spook as its national security advisor, and more.

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime for free every week, or level up here.


First up: Steady as they go

"Cattle, not pets" was a popular maxim within the budding DevOps movement more than a decade ago as the transition to cloud computing forced companies to realize that their distributed systems should be seen as replaceable parts, not members of the family. Fast forward to today, and Restate believes that if developers want to build applications around AI agents, they need to stop spending so much time building new backend logic for every app.

"It's a very generic problem: How do you build distributed backends in a way that is robust, reliable, that handles all the gazillion sort of things that can happen?" Restate co-founder Stephen Ewen told The Stack's Noah Bovenizer in this week's edition of STACKUP. The company just raised a $20 million Series A to build out what it thinks is "a fundamental infrastructure building block missing" from the development process.

STACKUP: The Stack’s weekly tech startups and funding wrap
This week’s rounds and milestones.

Their idea was that the concepts behind durable execution should be cheaper and easier to implement as agents start to take over enterprise workflows. "When durable execution becomes more efficient, applications use more of them and get simpler semantics and better debugging capabilities: durability inside the agent loop (persisting inference, tool calls, guardrail evaluations) ensures you always get a consistent execution history, compared to durability around the loop; that matters for agents performing sensitive, long-running, or expensive tasks," Ewen said in a blog post announcing the new funding.

As the DevOps movement learned, enterprise platform shifts often require a new way to think about application development in order to unlock the potential of the new platform. The complexity of running agents at scale — and the difficulty of monitoring their activity at scale — will require developers to build systems where durability and predictability were baked in from day one.


The Stack Summit: We're convening in London on November 4-5, for a series of exclusive workshops and fireside conversations on the rise of BYOC as a favoured enterprise SaaS deployment model; how many CDOs are consolidating their data estates with Apache Iceberg; how CISOs at FTSE 100 scale are handling supply chain risk (with GSK's CISO) and more. 

Ticket applications are subject to pre-vetting. Get in touch with ed@thestack.technology if you want to be in the room.

Learn more

The rest of The Stack

Four more years: The AI models released over the last year finally allowed enterprises to understand how AI agents could play a role in their businesses, but at some point the bottom line has to work out as well. Deutsche Telekom told investors Tuesday that it expects to see AI ROI just around the corner.

Deutsche Telekom says AI will save it €2.5bn by 2030
Europe’s largest telco is rolling out more AI tools “to be less dependent on people”.

AI in DC: Anthropic's relationship with the US government has been tested this year after the release of powerful new models, and it recently decided to bring on a new executive with significant experience in both the public and private sectors. Andrew Scott announced Monday that he'd joined the model company to work on "the intersection of cybersecurity and national security," following a one-year stint at Capital One and earlier gigs with CISA and the CIA.

Anthropic hires ex-CIA cyber chief to lead national security
“Today the question is different, and harder: what happens when AI makes cyber operations faster and cheaper for nation-states and criminals alike?”

Network effects: The Metropolitan Police Service wants to upgrade its aging networking infrastructure to a more modern system based around SASE, according to a proposal issued last month. That approach also ratchets up security protections around the network for the Met, which still relies on a lot of ancient gear.

Met Police eyes SASE with massive network overhaul
“Deliberate disaggregation of our infrastructure”

Heavy metal: Schneider Electric made the biggest deal in its storied history Monday, announcing plans to snap up PTC for $22.6 billion. PTC's industrial design software gives Schneider another software business to complement its earlier acquisition of Cognite and its own homegrown software tools..

Schneider Electric shells out $22 billion for PTC’s industrial AI software
The idea is to create a software portfolio that can allow AI agents to operate safely inside complex industrial systems.

Portal problems: Most travelers are familiar with the ritual of logging into public Wi-Fi hotspots through a portal before gaining full network access, but after warning about attacks that steal control of those portals earlier this year, Microsoft said Monday they're still happening. Windows users can try to limit their use of those portals to prevent attacks or implement an always-on VPN.

Microsoft flags renewed attacks on hospitality wifi networks
Compromised Wi-Fi login pages redirect victims to infrastructure controlled by Russia-linked attackers.

Stack ranking — Enterprise moves

Brendan Evers is the new CEO of Spur Intelligence, a promotion from his previous role as president and chief operating officer of the threat detection company.

Christian Werner is the new CEO of Xantaro Group, joining the IT services company from Digit Solutions.

Tarun Chopra is the new senior vice president and general manager of Intelligent Z Optimization and Transformation at BMC, joining the company's mainframe modernization group after several years at IBM.


We're also reading:

Tracing the rogue ideology of the frontier labs to their product choices: Sam Anthony, co-founder of Perceptive Automata, goes long on the history of LLM development and the concept of "alignment."

We’re going to need default hard budget caps on pretty much everything: FinOps wasn't a thing until later-arriving cloud customers were shocked by how quickly costs could pile up if you weren't careful, and Simon Willison explained why this is going to happen again with enterprise AI tools.


Thanks for reading — see you Thursday!

Add The Stack on Google