FastForward #85: Dispatches from the road
Hi everyone. It's November and it's getting dark early as the leaves turn brown and the sky turns grey. I’d like to thank Akamai for returning as a sponsor for the next four issues. If you're interested in advertising, please drop me a line at ads@fastforward.blog. If you like what you see, please share this week’s edition with a friend, and encourage them to subscribe. It really helps.💌 Sign up here.
ForwardThinking 🤔
Dispatches from the road
On the road again
I just can't wait to get on the road again
The life I love is making music with my friends
And I can't wait to get on the road again
~ Willie Nelson, On The Road Again
I’ve been doing a fair bit of traveling recently and I thought it might be interesting to catch you up on some of my observations from the road.
Last week I was in San Francisco, where I recorded a sponsored video interview with Egnyte CEO Vineet Jain and CTO Amrit Jassal, moderated a panel on agentic security for Realm Labs and attended two conferences: Constellation Research's Connected Enterprise and GitHub Universe. This week I was in Boston for an AWS media event.
Yes, it's been a busy time, but I’ve learned lots along the way about the realities and challenges of implementing agentic AI from basic change management issues and institutional inertia to the constantly-changing technology to the real challenges around agentic security and governance.
One of the ironies here is that as AWS executive in residence Tom Godden pointed out, yes, some companies have figured out how to build an agent, and they even have taken processes that took hours or days and cut them to a fraction of that, but it hasn’t necessarily produced better outcomes yet. There’s a kind of ‘what now and what did we actually gain’ kind of reaction.
In a concrete example, Godden pointed to a company he spoke to that reduced a process from two hours of work down to 35 seconds. The executive went looking for what that gained him and he struggled to find concrete metrics like more jobs completed or increased revenue. "The freed capacity we found without any destination is just overhead,” he said.
It reminds me of a couple of years ago when I was getting physical therapy for a problem I was having with my foot at the time. The therapist asked me to stand on one leg 30 seconds. I said I couldn’t do it on the bad foot. She looked at me and said, “you can't do it yet,” and she was right. A couple of weeks later I could. Large organizations are like me early in PT, it seems, they are struggling to find their balance.
Regardless of the event or the speakers, the themes were surprisingly similar. AI has the capacity to give you wings (to borrow a phrase from the Red Bull commercial), but that velocity produces a whole host of different problems and companies don't always know how to control their new super powers.
Speed creates its own set of problems
Coding is the most successful agentic use case yet, but just because you can produce code faster doesn't necessarily mean that you can keep up further down the pipeline, or keep product, sales, marketing and especially customers in the loop. That's the more difficult part.
As Egnyte CEO Vineet Jain told me, "Just slinging features very quickly doesn't get you any bonus points." Customers may lack the ability to pivot as quickly. They typically have configured the software and workflows to work a certain way within their particular environment, and they can't just slap on new features willy-nilly without testing and looking for possible issues. They have actual businesses to run. They aren't in the new feature implementation business.
Kyle Daigle, chief operating officer at GitHub, perhaps put it best when he said, "The [code] generation is easy, the creating and deploying software part is quite difficult." GitHub, of course, will be happy to help you with that, but the point is that the bottleneck is shifting from generating code to figuring out what to do with all the code that's being generated, and how you make sure it's safe, secure and compliant, and not just slop created by a bot.
And once deployed, especially with non-technical users involved, how do you keep everything safe and secure? As more agents get deployed, that challenge becomes more difficult. As Charles Iheagwara, head of AI cybersecurity delivery operations at AstraZeneca, said during a panel at CCE aptly titled Do you know what your agents did last night?, “Let's not confuse controls with observability.” He pointed out that vendors can provide elaborate dashboards full of metrics about what agents are doing, but how do you prove those metrics are accurate or that agents are operating within their prescribed boundaries? Being able to observe what agents do isn't necessarily the same as being able to control them.
All of these conversations point to a similar set of issues. AI can do some amazing things, but organizations are still struggling to figure out how to take advantage of it safely and securely. The trouble is that technology is moving faster than most companies can adapt, and if they can't figure out what to do with that newfound speed, they'll just end up creating a whole new set of problems faster than ever.
Disclosure: I was paid by Egnyte to conduct a video interview with its CEO and CTO and by Realm Labs to moderate a panel on agentic security during my visit to San Francisco. Neither company had any editorial input into this column.
~Ron
What's new on the blog 📰
Zurich US chief data officer recognizes that AI isn't the answer to every problem
I recently interviewed Kristen Bessette, chief data officer at Zurich North America. As an insurance company that deals primarily with business customers, it has to be particularly sensitive when it comes to introducing new technology like AI agents.
But Bessette said something super interesting during our conversation: The solution to every problem isn't necessarily AI.
"I told my team when I came in, listen we don't need a sledgehammer to kill a fly. Sometimes you just swat it, right? So if it's a swatable solution, just swat it."
How AI is rewriting the startup playbook
As AI speeds up getting that initial idea off the ground faster than ever, it raises questions about the role of the cloud provider in this new world.
I spoke to Darren Mowry, head of startups at Google Cloud, recently on the FastForward on PPN podcast about this changing role and how the cloud companies have to adapt quickly in a rapidly changing landscape. Mowry knows from whence he speaks having worked at Amazon and Microsoft in the past before joining Google Cloud 6 years ago.
As he told me, cloud credits are a given, but the cloud vendors have to move beyond that when it comes to working with startups. "We're all going to offer credits. Credits are not differentiating, right? The way that we help you get value from those credits is what's differentiating," he said.
The full podcast is embedded in the story.
AI velocity is still limited by human requirements in large companies
I produced a sponsored video with SailPoint's CTO Chandra Gnanasambandam last month during the SailPoint Navigate conference in Austin.
We discussed how any major change management exercise, and AI may be the biggest yet, requires companies to balance innovation with risk. Teams charged with protecting the company can slow the pace of change. Gnanasambandam says that keeping everyone moving forward requires constant communication and empathy.
"So the things that I talk about a lot these days is actually things around empathy, active listening. When you when you actually hear a legal colleague say something like, 'No, no, no, you can't do that,' I'm like, curb that knee-jerk reaction and you know assume positive intent and go back and ask a question why," he said.
Note that SailPoint paid me to produce this video in collaboration with their team, while I retained editorial independence over the questions I asked and final piece. If you want to me to do a sponsored video with your company, please feel free to reach out.
The 15 minute final interview is embedded in the post.
Qualcomm CIO Atilla Tinic is rethinking the IT playbook for the AI era
I spoke to Qualcomm CIO Atilla Tinic recently. We talked about how AI is forcing him to question long-held assumptions about how to run IT, including how companies use their data.
“I've done a 180 now, and I really believe your data lake or lakehouse is really your source of record for unified data going forward. And when you start to incorporate AI agents on top, they are more powerful in their capabilities when they have unified data that's trusted and curated that they can leverage.”
Your next tough work conversation could start with an AI avatar
I spoke to a couple of Synthesia execs recently about how they are using avatars to teach managers how to have tough conversations with employees. It lets them practice these conversations in a safe way before they meet with the employee.
Read the full story>>
News of the Week 📣
CrowdStrike's new AI kill switch is really an agent access cutoff button
As fears spread that unfettered AI could cause systemic damage, the concept of an AI kill switch has emerged. This is the idea, that you hit the big red button (metaphorically at least) and it shuts down the system.
Such an uber button is probably not realistic, but that didn't stop CrowdStrike from announcing an AI kill switch this week at Fal.Con Europe in Barcelona, and making the initial functionality available immediately.
A more apt name, however, might be a rogue agent kill switch. Let's suppose you observe an agent doing something it's not supposed to do such as, say, I don't know, just riffing here, hacking Hugging Face, you could implement this kill switch and cut off access to the underlying model, effectively stopping the agent.
As CrowdStrike describes it, "If an agent breaks out of its intended behavior, a defender can terminate it where it runs, block it from launching again, and cut off its enterprise network access."
For now, that involves blocking an agent's access to the model through CrowdStrike's AI gateway or a third-party gateway. Future capabilities will allow broader agent containment across CrowdStrike's Falcon platform.
It might not stop the underlying models, but at least it can put down a misbehaving agent, and that's a start.
Marvell makes a $115 billion bet on optical technology with Lumentum deal
Chip designer Marvell announced this week that it has agreed to acquire Lumentum, a maker of light-transmitting optical hardware, in a $115 billion all-stock deal including debt. The move gives Marvell technology and manufacturing capacity critical to moving data through AI data centers.
Marvell and other industry experts believe copper wiring is reaching its limits as AI workloads grow, forcing a shift toward optical connections. "Surging AI data traffic requires rapid hardware innovation, and this combination enhances our ability to deliver next-generation solutions at scale," Lumentum CEO and president Michael Hurlston said in a prepared statement.
Patrick Moorhead, founder and principal analyst at Moor Insights & Strategies, thinks Marvell is well suited to make the most of this deal. "Marvell is really good at choosing and integrating other companies. They have a great track record on this," Moorhead told FastForward. "It’s expensive but Lumentum brings a lot to the table that Marvell does not have. It indicates to me that valuations and opportunities are high and that more integrated solutions versus piece parts are desired."
In a LinkedIn post, Moorhead said he believes the combined companies are going to solve a big problem. "The copper wall is real. I said in March that scale-out and scale-across were already optical and scale-up was next. When the link goes optical inside the rack, the DSP, the laser and the packaging have to be designed as one thing. That's the whole thesis, and it holds."
Remarkably, Marvell announced it doesn't expect the deal to close until the first half of 2028. That's not a typo, that's somewhere between 14 and 20 months from now. That's a heck of a long time for a deal to simmer, but perhaps the companies expect a lengthy regulatory review.
Box adds AI workflow design to its automation toolkit
Box is no stranger to workflow automation. It has been incorporating workflow tools into its platform for over a decade going back to Box Relay. This week, it announced the latest iteration, a new tool called Box Architect that lets users describe what the workflow should do, and it goes and builds it.
Box sees the advances being made across the model landscape and is trying to build a tool to take advantage of that and simplify the act of creating workflows. Instead of using low code or visual tools as in the past, with Box Architect you simply tell the model what you want it to do, point it to the right content and it will build an entire workflow for you, including the necessary agents, apps, automations and metadata templates (at least in theory).
Nirmal Ganesh, Box’s senior director of product management for Box Automate, described the tool on stage at BoxWorks this week as "an autonomous agentic system that partners with your enterprise to reimagine workflows for you."
Box is hoping to take workflow creation to the next level, taking advantage of the latest capabilities in the frontier models. While the onstage demo made it look like it's presto-chango, easy peasy, in reality, building a complex workflow like the ones Box described this week takes a fair bit of human tweaking and back and forth to get to a steady state. The tool is built to design, test and deploy and the testing may take a fair bit of time to get right.
And with the models constantly changing, getting there today doesn't mean it will work tomorrow. So while Box Architect sounds like a big step forward, it's important to remember that even with the help of vendors like Box, companies continue to struggle to build agents that deliver consistent and meaningful results (see this week's ForwardThinking commentary).
Box Architect is not generally available yet. It will be launched in beta in Q4, so soon, but at least initially, this tool will only be offered to Enterprise Advanced customers.
What I'm reading 📚
He Trolls AI Labs for a Living. Now They All Want to Work With Him
~By Isabella Ward, Wired
“I’d Know That Voice Anywhere”: Inside the €95 Million Fraud That Brought an Italian Bank to Its Knees
By Keegan Henckel-Miller, DeleteMe blog
How the A.I. Boom Is Driving Up Sales of Gray Paint
~By Emily Martin, New York Times
What I'm watching 📺
Why Sharks Attack The Internet
~The Rest is Science with Hannah Fry and Michael Stevens
Look who's talking 👄
"But adoption only tells you if people open the tool. Almost no one we talked to could answer the questions underneath it: Is it producing better work? One organization that we spoke to told us that they had 88 percent adoption of AI across their company, but they also told us that they estimated fewer than one in 5000 sessions with AI actually produce better work. "
~AWS executive in residence Tom Godden at AWS Media Day this week
