The state of AI in 2026: On the road to ROI(mckinsey.com) |
The state of AI in 2026: On the road to ROI(mckinsey.com) |
People think they can do it themself... kind of like how everyone build their own database using Microsoft Access in the 90's. When all software developers needed to repair all the issues...
Plenty of companies run most of their internal operations on Excel and Access for exactly this reason. And most developers barely know anything about databases anyway.
It is far more plausible that we'll see that pattern again - small, department-level vibe coded apps will be around. And things that need to scale bigger will need professionals.
I suspect the "or features" there is doing a lot of work.
I kid of course - their recommendations on management of technical teams are fly-in, fly-out out of context garbage in my experience. Usually results in poor but justified decisions from senior leaders that happen in cycles of 3-5 years.
You have to recognize that a vast majority of enterprises fall in the category of "Dumb IT". To them, IT is a cost-center and overhead as their main line of business is something else. And, they look up to the likes of McKinsey for wisdom.
I get that these companies (airlines, widgets, whatever) don’t see IT as a thing that generates revenue, but it’s like having electricity or running water. You don’t think about it much, but if you don’t have it, you’re fucked.
I view this mostly as a case of uneven diffusion. Consider your priors on the tech savviness of the bottom 50% of enterprises, let alone the bottom 25%. Many companies won’t adopt AI for 5 years. Now model how a 5% difference in growth plays out; not enough to sink stragglers immediately.
The other interesting bit is a 5% increase in EBIT could be a smaller productivity increase in the margin, eg you lowered prices by 2% and some fraction of your competitors’ customers noticed.
Basically, nothing revolutionary but I think it’s an error to read this as “we don’t see ROI yet”.
The other interesting point is one of legibility; the difference between “employees using AI” and “business function has rolled out AI”.
> Among AI tools, chatbots are the most widely scaled, with 47 percent of respondents saying their organizations are scaling them across the enterprise
We should look into the details, but I read this as: ~everyone is using AI ad-hoc, the 47% is the consultant-y “leadership has stood up a measurable initiative to scale AI…”. By construction, Gartner cannot measure the on-the-ground adoption.
Do the most unethical profit-maximizing thing, but also account for the eventual lawsuits and loss of goodwill in your calculation.
(Statement has been made since 2022)
The thing I think about is AI is already at that spot where you don’t need a survey or a Randomised Controlled Trial to prove it makes one productive. Sometimes things are that obvious. But these kind of tests and surveys appeal to a certain kind of person who considers IMO the wrong type of things as rigorous.
I think companies will get better at measuring the impact of AI and attributing it to increasing profit or decreasing costs, and that the companies that are better at this will have an advantage over those that are worse, so eventually overall efficiency will improve.
There exists no evidence that manually trying to fix or rewrite the system is going to be more efficient today.
And if AI improves even half as fast as it did over the last year, all bets are off for what 2028 will look like.
So rather than debug a steaming pile of code you might be be able to treat it as a greenfield opportunity.
No, thank you. I’ll leave the underpaid disinfestation job to someone else.
The best fun is when you’ll be called in not to untangle the LLM mess, but to build new features on top of its nonsense architecture and API design, but expected to do it by hand because they have lost confidence in AI.
Also the Y2K folks made bank IIRC.
The same time many of the overnight jobs started up :-)
Fortunately we had zero issues.
If an implementation is "a black box that works," absolutely, LLMs are fine. They can create complex programs in full.
But if one expects a tight implementation... it's a horror. LLMs vastly overengineer code (I believe they're intentionally designed to do so). I really struggle to make LLMs generate lean code.
In the best case, they generate bloated designs (in terms of complexity, not necessarily performance), but in the worst case, they generate gaps in the specification. And even in the first case - cognitive load is a problem also for LLMs, not just for humans (although obviously at a larger scale).
I think LLMs are essentially modern compilers, with similar problems, but designed more like "deoptimizing compilers" than optimizing ones. :)