FastForward #71: The hardest part of enterprise AI isn't the AI
Hi everyone. I've been on vacation this month, but I had an itch to write, so I decided to send out an abbreviated newsletter this week without news of the week. If you like my newsletter, please share this week’s edition with a friend, and encourage them to subscribe, it really helps.💌 Sign up here.
ForwardThinking 🤔
The hardest part of enterprise AI isn't the AI
I've been off for a couple of weeks, enjoying vacation, but still keeping one eye on tech news, and it's a little unbelievable how much happened in two weeks in July, a time when news is supposed to slow down.
Consider that over the last couple of weeks:
- We had major model releases from OpenAI and Anthropic.
- Microsoft launched the Microsoft Frontier Company to embed engineers inside customers.
- Meta increased its data center investment in Louisiana to $50 billion.
Meanwhile, Mira Murati's company, Thinking Machines, released Inkling, a customizable model aimed at enterprises that continue to struggle to adapt general-purpose models from the big AI labs to their own needs.
According to Axios reporting, the startup hopes that by developing a more flexible model, customers can get better results at a lower cost. It's certainly a creative approach, but it's not yet clear whether it will solve the fundamental organizational problems that appear to be at the root of companies' AI challenges.
It ain't easy
When you look at Thinking Machines in the context of these other major announcements, a story begins to unfold. Microsoft, for example, has chosen to deploy (expensive) engineers inside customer companies to help them find more success implementing AI. OpenAI announced a similar approach in May. Both launches represent a tacit admission that, surprise, surprise, AI is hard to do well.
Perhaps the reason Microsoft decided to launch a new organization to help customers implement AI is because research shows that most companies continue to have problems taking advantage of AI, even as the models become more capable.
A smarter model doesn't help you get good data to feed the model. It doesn't help you rethink workflows or change the way you work. All of these things require understanding most companies don't have, and consultants in theory should be able to point them in the right direction -- for a price of course.
Thinking Machines is trying a different approach. Instead of building the most powerful model, which is great for bragging rights, but not necessarily practical, it's shooting for something that enterprises can work with. Whether Inkling is the solution or not, it's an acknowledgement that companies are having a hard time and they need a model that helps them work smarter, not necessarily the smartest model.
The struggle is real
The Thinking Machines approach might work better if employees weren't using AI. But the research suggests that they actually are. The problem is they don't appear to be getting the most out of it from an organizational perspective, an important distinction.
After interviewing 750 employees and leaders across industries, a July 2026 McKinsey Report found that "Most organizations are still early in their AI transformation journey—and that employees are more ready to use AI than their organizations are to change around it." The fact that we are almost four years from the launch of ChatGPT, and companies continue to struggle in spite of internal usage is telling.
And executives are getting impatient. A Bain report released this month that asked 100 CEOs about their company's AI efforts found that "roughly 80% said they are dissatisfied with the progress of their AI programs, despite the heavy investment and high expectations." That's probably because they aren't getting the promised efficiency gains, while spending increasingly large sums of money trying to get there. That's not a combination that is going to make many CEOs happy.
Both reports suggest there is a fundamental organizational disconnect between AI usage and getting the most out of the technology. The Thinking Machines solution is to solve this with a more flexible model, but that still doesn't fix the basic organizational problems.
As we have learned with other major technological shifts, you have to change more than the software to get the most out of the transformation. That usually requires additional investment and a long-term and focused commitment companies have a hard time sticking to, especially when the cost of doing business keeps going up without good results.
~Ron
What's new on the blog 📰
Investor Jai Das on AI hype and the changing face of enterprise investing
I caught up with Sapphire Ventures co-founder Jai Das in April at HumanX. We talked about AI hype and the changing investment environment.
As the AI labs try to muscle into areas traditionally covered by Enterprise SaaS companies, he says there are still areas where enterprise investors can succeed.
"I'm in a little bit a better position than my colleagues who look more at apps because I do more on the infrastructure and software. So those layers, I think, are not going to get disintermediated by the models as much.”
Why Abby Kearns thinks AI has changed the rules for the open source software supply chain
I first met Abby Kearns when she was running Cloud Foundry a decade ago. Today, she is CEO at ActiveState, a company that is trying to create a way keep software libraries safe at a time when the open source software supply chain has been under increasing attack.
She said recent moves by the Linux Foundation and IBM and Red Hat show the industry recognizes there is a serious problem.
"When institutions of that scale stand up coordinated programs around open source supply chain protection, the consensus has changed. This is no longer treated as a niche security topic. It is a crisis and recognition is finally catching up to the pace at which the exposure has been building."
Databricks continues to soar as private company, but margins starting to shrink
In this the latest collab between me and Alex Wilhelm, we dive deep into the success of Databricks, which continues to soar as a private company. Alex and I have been covering this company for a long time – my favorite part of the article is where I link to like a half dozen analysis pieces we wrote together over the years when we were at TechCrunch. I wasn't kidding when I said we've covered them a long time.
This will be cross-posted on Alex's Cautious Optimism blog.
Zscaler CEO Jay Chaudhry's latest bet focuses on protecting enterprises from AI-fueled threats
Speaking to Zscaler founder and CEO Jay Chaudhry, it's hard not to be impressed, even if as a journalist I'm supposed to be above such things.
From his back story growing up in a small village in India to his move to the U.S. with a one-way ticket and $200 in his pocket to starting 5 companies, four of which were acquired before launching Zscaler in 2007, he has had a long and successful run, and he shows no signs of slowing down.
I spoke to him across two conversations. One in April and one at the company conference in Las Vegas earlier this month where he talked about going after the threat AI adds to the security equation:
"While these models can generate code, they can also create exploits, and they can create a fix for the exploit," Chaudhry said. "Here's the difference: If a model created an exploit, and it only works 90% of the time, the attacker doesn't care. If your fix only works 90% of the time, you have a problem."
Note that I had a paid engagement with Zscaler when I attended their conference. My editorial standards and guidelines apply regardless. I have a full disclosure statement at the top of the piece.
FastForward on PPN podcast featuring Don Schuerman, CTO and head of marketing at Pega
In our latest episode of the FastForward on PPN, I spoke to Don Schuerman from Pega Systems. We talked about his longevity, spending 28 years at one company, the ever evolving market, the implications of content velocity, the company's recently announced outcome-based pricing and much more.
What I'm reading 📚
AI FOMO Isn't a Strategy
~By Malvika Jethmalani, Reworked
AI is stealing all the RAM and storage, and I’m learning to live with it
~By Harry McCracken, FastCompany
I got slopped
~By Kashmir Hill, New York Times
What I'm watching 📺
Why Most Startups Shouldn't Raise Money (Even If They Can)
~Walter Thompson, Fund/Build/Scale
Look who's talking 👄
"Mr. Johns reads what ChatGPT produces, pushes back if he thinks it needs changes and then copies and pastes the text into a Word doc. He aims for 10 to 15 chapters, which work out to around 300 pages, and then uses Amazon’s Kindle Create app to convert the text into an e-book manuscript. He has ChatGPT generate a cover image — generally vastly superior to the plain one on my biography — and then uploads it all to Amazon, which does not charge for the privilege. His goal has been to write 10 books per week, which is the maximum Amazon allows."
~Kashmir Hill, I got slopped, New York Times.
