So in 1987 when Walmart completed the largest private satellite network in the world, the man was sweating the $24 million price tag ($70.4 million today).

The satellites linked every store to the Bentonville HQ and looked like a terrible call when the buggy system’s payoff wasn’t visible in any quarter.

Suddenly, the satellite network made it possible to compile same day sales data from every store. The supplier partnerships boomed, and Walmart knocked Kmart out of the retail game forever.

The invisible infrastructure to make this possible arrived years before the benefits, and led Walmart to retail dominance. The same thing is happening in 2026.

Sparky?

Walmart went all-in on AI agents.

They actually call them “super agents”, which causes the internet’s collective eyes to roll. BUT in typical Walmart fashion, they do not take emerging tech investment lightly.

This became vivid on the August, 2026 Walmart earnings call, when CEO John Furner tells us Walmart’s super agent “Sparky” has seen usage rise 70% year over year.

You might be thinking how trivial this is considering the proliferation of AI generally. Everyone is using ChatGPT, so who cares that Walmart has AI tools you can access on their app?

But Furner went on to drop some data that made the AI bears recoil, saying customers who use Sparky spend 40% more per order at Walmart. Everyone’s minds are blown by this considering the amount of people preaching doom on AI spend.

The evidence for what that AI money actually did comes later in the same call, when Rainey reaches SG&A and says Walmart leveraged wages in the quarter.

He is using a term that retail accountants love to confuse the masses with. In plain terms: payroll grew slower than sales.

Walmart’s labor as a percentage of revenue fell, which is an incredible byproduct of employees in the stores wielding AI tools that Walmart already bought. Supply chain automation also streamlines the flow of inventory into stores as well, showing boosts to this same wage leverage Rainey mentioned.

The catch comes in the next part, when Rainey says higher depreciation on that same capital expenditure, along with rising self-insurance costs for Walmart, more than offset the wage benefits they gained. Depreciation grew 12.6% in a quarter when sales only grew 5.9%, showing how depreciation can take a toll in the short term.

While Sparky is leading to higher spend by customers that is correlational in nature, it shows that consumers are getting benefit from AI agents in a retail environment.

The deeper discussion on leveraging wages is the way in which Walmart will eventually see major ROI from AI spend. This quarter it didn’t work in Walmart’s favor, but don’t have an anti AI crash out yet.

The same workforce absorbing $100 billion of additional revenue since FY23 will be the main payoff, rather than what everyone else assumed would happen when AI proliferates throughout the organization.

We have a narrative violation, as people thought savings would come through AI causing headcount reduction through layoffs, but this just isn’t playing out. Further, the depreciation point by Rainey tells us the equipment enabling this AI transformation costs more right now than the labor efficiency is worth.

We are at the not so fun part of the investment j-curve where things look grim, but Walmart is making huge bets on AI. While the automation made Walmart more efficient and less profitable than they could have been at the same time, they are also starting to win with AI investment. This is just step one.

We should ask what foundations Walmart had to build before gen AI could do anything useful for their business, customers, and shareholders and how can we use these foundations to propel our own companies forward in 2026?

Subscribe now

The Golden Foundation

Walmart’s Sparky is an agentic interface living on top of an inventory system that already has a brilliant organizational design.

This infrastructure enabled Sparky to know where things are located, and this is the win for consumers. The CEO’s Sparky anecdote from the earnings call proves this when he shows how a customer asked Sparky for a high-protein meal plan.

Going above and beyond what a normal LLM like ChatGPT would have been able to deliver, Walmart’s Sparky builds the components of a meal plan with its agentic capabilities in the Walmart shopping cart. Sparky accomplishes this with a “one click add” for each of the ingredients that saves the user from having to find all of the items themselves.

It does this while skipping ingredients the customer has recently purchased both in store and online because it knows they are irrelevant. While frontier AI models could build meal plans for years, Sparky uses a foundational layer of infrastructure to actually help the customer navigate online and lower the time to checkout.

What consumed nearly 10 years of effort for teams at Walmart is laying underneath the agentic AI layer of Sparky, and these are systems that allow the AI agent to deliver something invaluable to the customer rather than just be pure hype for investors.

For its novel features, Sparky grabs the reins of the unified transaction data across channels, products, and real-time availability of the ingredients the customer would need for their meal plan.

This was just one example in the earnings call of how the play to combine infrastructure + the AI layer is the star for how we ascend in deploying AI.

Here are two more:

• 3,100 US Walmart retail locations now use some level of automated freight handling systems, boosting supply chain efficiency to new levels.

• Over 50% of e-commerce fulfillment is flowing through automated facilities where AI helps workers get customers their shipments faster.

The supply chain layer underneath didn’t just spawn overnight. For example, Walmart has been forecasting for their supply chains on a multi-horizon recurrent neural network built by their own engineers for years. It’s a system that predicts several time distances simultaneously, because different decisions with your supply chain will have different lead times and leadership MUST be able to see this clearly.

They also employ a digital twin as a simulation of the physical network of their supply chains and no this isn’t a weird AI clone that lives in the cloud. This digital twin is like a flight simulator sandbox with stores, distribution centers, trucks, and inventory that lets Walmart test changes before implementation (drive it like they stole it).

They also apply intelligent computer vision on the task of inbound quality checks, so cameras inspect produce as it comes off the truck.

The purpose of this tech is grading freshness on perishable goods and catching damage that is too time consuming for humans to perform. Image classification of this caliber has been viable since about 2016 in receiving environments, so the infrastructure has been marinating for a while.

The supply chain automation technologies noted here predate ChatGPT by nearly a decade or more. This hits hard because most of what people credit to Walmart’s gen AI investment is actually classical machine learning. This confusion happens a ton in 2026.

The thing to remember is that gen AI is the routing and interface layer on top that gives customers access to everything Walmart has been building for years in relative silence.

When a customer asks Sparky for a week of high-protein dinners and it skips the ingredients they already bought, the language model handles the dialogue as the most futuristic interface we currently have.

Every fact underneath comes from the forecasting, inventory, and transaction systems that took 10+ years to build when the engineers were in monk mode.

This is the advantage that legacy companies need to be giving themselves before they dive head first into agentic AI offerings for their customers. Plant today.

Subscribe now

Are We Just Renting The Front Door Now?

Distribution through someone else’s interface is fine, while transactions and customer data through someone else’s rails is not going to work out long-term because you don’t own the most valuable part. Walmart proved this in 2026.

Back in October 2025, Walmart and OpenAI announce an Instant Checkout feature inside ChatGPT which blows everyone’s minds at the time. However, by March 2026, Walmart scraps the idea and makes new plans.

The real power implementation is Sparky itself embedded inside ChatGPT, where customers authenticate first with their Walmart account. Amazingly, their shopping carts then sync across the app and site, and payment runs through Walmart’s own checkout system.

According to Daniel Danker, Executive VP of AI Acceleration, Product, and Design at Walmart, the mission of this implementation is to ensure shopping can start anywhere that customers get the urge to buy through research, and the Walmart experience stays the same.

The distribution layer does the heavy lifting, as Danker says ChatGPT alone brings Walmart new customers at about twice the rate of traditional search engines. This is likely because ChatGPT’s users skew away from Walmart’s typical shopper and they are capturing entirely new segments.

The transaction layer is what failed for the mission and caused Walmart to pivot, because they had a sudden enlightenment where they realized the transaction layer is king.

Walmart starts this fail by putting around 200,000 products into OpenAI’s Instant Checkout system starting in November of 2025, so people can purchase popular products without even leaving the chat.

This doesn’t work as Danker told WIRED that purchases converted inside of ChatGPT are three times lower than letting the same shoppers launch to Walmart’s own site for checkout. The fix is to stop outsourcing the transaction to OpenAI ‘s platform and risking the loss of sales while keeping the ChatGPT distribution channel open to bring in new customers now that traditional search engines are on the way out.

Every legacy company is about to face a version of this with the proliferation of customers using LLMs for search, and operators need to be ready. Walmart’s data on the distribution side is too enticing.

When a vendor offers to build out your product catalog, your quoting system, or your booking process inside their assistant, be ready with this viewpoint from Walmart. Let the vendor handle the distribution channel while your teams keep the account, the transaction, and the records for you to leverage on your own.

One Platform, Many AI Models, and A Few Good Agents

It is evident that the real scarce asset is a governed place to plug in LLMs, plus the rare discipline to know how many good interfaces you ship.

By a place to plug in, we are talking about Walmart’s Element.

This is Walmart’s internal multi-cloud platform that allows teams to connect to multiple managed LLMs safely. This platform even routes to the models by cost and performance, and carries the security review and governance layer Walmart needs in order to let employees loose on LLMs without risk of doom.

Walmart’s Global CTO Suresh Kumar said to IT Brew this June that the way to drive governance is to just lock in and build it (into the tool).

To follow this mission, Walmart bizarrely names a tool called Code Puppy. This is because it is an internal vibe-coding system that lets all types of employees build software tools. However, it only talks to LLMs that Element has already vetted before allowing teams to go insane vibe-coding slop on just any models out there.

The result Kumar reports is that Walmart now has roughly as many non-technical employees building things as it has engineers, boosting their coding workforce from prior years before vibe-coding was possible.

Next, consolidation is also important in this context. People might think of Walmart as this legacy company that moves slow, but even by mid-2025 Walmart has 200-plus agents in use across its business divisions. They deploy so many agents they build another tool called Wibey to coordinate them internally without spiraling.

The most impressive part of this endeavor is that Walmart builds a small set of super agents based on the intended audience for each. Sparky was built for customers as we already mentioned, but they also made Marty for sellers and advertisers, plus one for associates, and one for developers to manage their respective domains.

Depending on the force of your company, you probably cannot build Element tomorrow, and you don’t even need to. The part for operators in legacy businesses to remember is the shape: you just need one approved way to reach AI that is safe and effective.

In order to accomplish this, you need one person who owns the system, and a limit on how many AI assistants and tools that employees are asked to study (this is where people jet).

Most companies fail the second one, because every department buys its own tools, causing total chaos that easily could have plagued Walmart if they didn’t think so precisely about AI implementation.

Subscribe now

The Real Score Card to Consider

This is where we see Walmart’s real AI metrics come alive as P&L lines on the statement, and the impressive ones have some serious caveats we will discuss.

As we address, the fact that customers who use Sparky order 40% more per checkout is correlational, but still cool. Sparky users self-select, and they are already serial app-native shoppers who are Walmart super fans.

The interesting part is that number is just 35% two quarters ago and jumps to 40%, which could mean either the Sparky product improves or the user dataset changes. So how do we know an AI tool that we invest in heavily actually changes anything if all we see is correlation for the 40% figure?

When we dive into investigating a major change like this we look for contrast.

Walmart’s engineers already know how to answer, because they run the tests properly one layer down to get an accurate reading (copy this playbook). A Walmart Global Tech paper published last year describes a two-week A/B test of a new search ranker on Walmart’s website.

The engineers split website traffic, run the test, and measure add-to-cart rate per each shopping session rising 0.8% total. The change is small, but after this test they know the tool causes it, because the only difference between the two groups is the tool being available.

In the end Walmart beat expectations and the stock still drops about 9% after the reports hit the media, because investors see US comparable sales come in at 2.6% against consensus near 3.5% to 3.8%.

A feel good AI story about Sparky and AI agent implementation does not move the stock price as much as people might expect in the “AI bubble”. The quarter is just soft, even with AI glow added.

For operators inside of legacy companies, this means you need to look at AI spend against a line finance already tracks so that you actually get eyeballs on your returns. This means labor as a percentage of revenue, rework hours, or even cycle time from request to delivery so that you know what is truly impacting your service.

Just measuring adoption rates and license counts inside of your organization are activity metrics that will get your project shot down. Fixating on adoption is what gets an AI program axed after two years because nobody cares about how many people are using an AI tool if it doesn’t influence P&L lines.

Your team also needs to run some kind of control group, however crude, so you have meaningful data to prove what you’re building is worth it, just like Walmart engineers do for Sparky.

The Hunt For Next March

We just spent paragraphs proving Walmart’s AI story is uncommon because it can be fully tested and graded by the outside observer, while also leading us to know how to lock in this year and deploy agents ourselves in an enterprise environment. Focus on the foundation layers as the primary.

We also realize that most of what legacy companies say about AI implementation on their earnings calls in 2026 is unfalsifiable by nature, which is why they yap about it so much to influence investors without much worry of analysis.

The future 10-Ks for Walmart will state a total associate count, which will give us the ammo to evaluate their AI implementation success and whether the current trends continue.

These future quarterly filings will say whether operating expenses levered or became deleveraged and none of this can be buried in hype language. Maybe it’s data like this that will implode the alleged “AI bubble” so many claim to be rising.

BUT If total associate headcount at Walmart stays near 2.1 million while revenue keeps climbing by next March, and labor productivity keeps going up, the mechanism could be real and the mission blueprint is worth taking for your own organization.

Keep building the foundation either way.

Subscribe now

Works Cited

“AI Shopping: Walmart and Gap Split over ChatGPT and Gemini Checkout.” Axios, 24 Mar. 2026, www.axios.com/2026/03/24/ai-shopping-walmart-gap-chatgpt-gemini-checkout.

“Before the Storm: Walmart Supply Chain Technology Helps Teams Prepare for Severe Weather.” Walmart Global Tech, June 2026, tech.walmart.com/content/walmart-global-tech/en_us/blog/post/before-the-storm-walmart-supply-chain-technology-helps-teams-prepare-for-severe-weather.html.

“4 Ways Walmart Is Scaling AI to Unify Its Supply Chain.” Supply Chain Dive, www.supplychaindive.com/news/4-walmart-supply-chain-ai-uses/760891/.

“From Models to Agents.” Walmart Global Tech, public.walmart.com/content/walmart-global-tech/en_us/blog/post/wibey-announcement.html.

“How Walmart Is Securing a New Cohort of AI Builders.” IT Brew, June 2026, www.itbrew.com/stories/how-walmart-is-securing-a-new-cohort-of-ai-builders.

Kumar, Suresh. “All in on Agents.” Walmart Global Tech, 24 July 2025, tech.walmart.com/content/walmart-global-tech/en_us/blog/post/all-in-on-agents.html.

Shang, Hongwei, et al. “Knowledge Distillation for Enhancing Walmart E-commerce Search Relevance Using Large Language Models.” Companion Proceedings of the ACM on Web Conference 2025, Association for Computing Machinery, 2025, pp. 449–57, doi.org/10.1145/3701716.3715242.

“Walmart and OpenAI Are Shaking Up Their Agentic Shopping Deal.” Wired, 18 Mar. 2026, www.wired.com/story/ai-lab-walmart-openai-shaking-up-agentic-shopping-deal/.

“Walmart Bets on AI and Digital Twins to Shape Its Supply Chain Strategy.” Retail Dive, July 2026, www.retaildive.com/news/walmart-supply-chain-strategy-ai-digital-twins/825162/.

“Walmart CEO: AI Is ‘Literally Going to Change Every Job.’” CNBC, 29 Sept. 2025, www.cnbc.com/2025/09/29/walmart-ceo-ai-is-literally-going-to-change-every-job.html.

Walmart Inc. Annual Report (Form 10-K) for the Fiscal Year Ended 31 January 2026. U.S. Securities and Exchange Commission, 13 Mar. 2026, www.sec.gov/Archives/edgar/data/104169/000010416926000055/wmt-20260131.htm.

Walmart Inc. Current Report (Form 8-K): Chief Executive Officer Succession Announcement. U.S. Securities and Exchange Commission, 14 Nov. 2025, www.sec.gov/Archives/edgar/data/104169/000010416925000172/pressrelease111425.htm.

Walmart Inc. Current Report (Form 8-K), Exhibit 99.1: Second Quarter Fiscal 2027 Earnings Release. U.S. Securities and Exchange Commission, 20 Aug. 2026, www.sec.gov/Archives/edgar/data/104169/000010416926000145/earningsreleasefy27q2.htm.

Walmart Inc. Current Report (Form 8-K), Exhibit 99.2: Second Quarter Fiscal 2027 Financial Presentation. U.S. Securities and Exchange Commission, 20 Aug. 2026, www.sec.gov/Archives/edgar/data/0000104169/000010416926000145/earningspresentationfy27.htm.

“Walmart Reveals Plan for Scaling Artificial Intelligence, Generative AI, Augmented Reality and Immersive Commerce Experiences.” Walmart Corporate Newsroom, 9 Oct. 2024, corporate.walmart.com/news/2024/10/09/walmart-reveals-plan-for-scaling-artificial-intelligence-generative-ai-augmented-reality-and-immersive-commerce-experiences.

“Walmart’s AI Assistant Sparky and Customer Spending.” Moneywise, Aug. 2026, moneywise.com/news/top-stories/walmart-ai-assistant-sparky-customer-spending.

“Walmart Sees Speed, Convenience Boosting Trust in AI Agent.” Retail Dive, www.retaildive.com/news/walmart-sees-speed-convenience-boosting-trust-ai-agent/813111/.

“Walmart (WMT) Q2 2027 Earnings Call Transcript.” The Motley Fool, 27 Aug. 2026, www.fool.com/earnings/call-transcripts/2026/08/27/walmart-wmt-q2-2027-earnings-call-transcript/.

This goes beyond necessity being the mother of all invention, as emergencies in business will often require executives to make huge bets on novel technology.

This is often technology that is unproven in the arena, and sometimes it implodes, but sometimes it works so beautifully that an entire company is saved from the probability of doom reaching 100.

In this inspiring case of a corporate house fire, it was a recently glued together house, still oozing at the seams after the Warner Bros. and Discovery merger in 2022. This is when executives made huge bets on generative AI in the midst of a writer’s strike and numerous other storms.

WBD was not a company full of merger novices by any sense.

In fact, WBD’s CIO Dave Duvall says WBD was “a nesting doll of past transactions, with each integration getting to about 80% before the next one began. The fragmented data and platforms make scaling an AI initiative much harder than it might appear.”

In spite of these challenges, WBD executives bet on AI’s ability to provide the leverage to wedge the company out of a bind on multiple fronts and into the colosseum limelight that is the brutal media landscape of the 2020s.

In a recent piece for MIT Sloan Management Review, authors George Westerman and David Kiron illustrate with great detail how WBD pulled this off.

In fact, they show us how a legacy company’s first two years with AI are spent taking inventory of its own potential.

Rather than a simple inventory of the technology available and its various strategic campaigns, it is an inventory of its own content (for data), workflows (for leverage points), and the limits when creative people will consider handing over their most beloved processes to the machines.

Subscribe now

Where It All Began

WBD was seven months into their most recent merger with Discovery when layoffs were ravaging the company as they struggled to get the culture right after blending two very different media giants.

On top of that struggle, strikes were heating up in Hollywood with Writers’ Guilds and Screen Actors’ Guilds members walking out protesting AI creep around 2023, in part due to AI models threatening their jobs that were trained without permission, often on the very types of content WBD owns.

The obvious play for leadership within WBD would have been to take the easy route and push AI heavily as an internal tool to cut costs so the leadership teams could breathe.

The gen AI tools and the models that supported them were getting so good during this time with the rise of ChatGPT post 2022 that adoption turned out to be the easy part.

The use cases were everywhere. The difficult play was fixing the data sewn beneath their surface and the fear flying over their heads, and that focused strategy is why the program survived at all.

There were tons of efficiency initiatives, but what the company lacked was a mission that could re-vitalize the executive team and a creative workforce struggling to pickaxe through a mountain of cost initiatives.

This was happening at the exact moment that the entire media industry was in a philosophical civil war of futurism regarding whether gen AI would displace creative jobs.

What happened next was a perfect example of how gen AI should be positioned: as a growth engine for creators. Many companies in this situation would have positioned it as something being used to cut costs for a workforce that had already gone through two rounds of consolidation, and it would have been ruinous for the culture.

Rebecca Kent, the head of transformation for WBD at the time, played this strategy perfectly as a way to implement AI in the right framing that would inspire employees to find the best use cases to create new IP for WBD, rather than be afraid of it as a cost cutting measure meant to destroy their roles.

They got the creative teams to go as all in as possible on using AI to improve the product without giving up the most treasured parts of the artistic process.

The second decision here was a perfect example of what we also read in McKinsey’s AI Transformation Manifesto a few weeks ago, that business leaders need to own the initiatives with IT serving as the support structure.

Rebecca Kent’s transformation team, boosted by the strength of Matt Chun’s corporate strategy group owned the initiative themselves while IT provided support based on the tooling and architecture needed.

Many leaders in legacy companies today have this totally inverted. While their IT teams nerd out on the newest AI capabilities, company leadership fails to make the jet go supersonic.

Another congruent theme we discussed recently that was also outlined in McKinsey’s AI Transformation Manifesto, appears with how the initiatives to engage with AI tools started at the top of the organization.

Company leaders worked through building a marketing case end to end, complete even with gen AI assisted jingles they created.

It’s hard to believe gen AI tools at this time were producing marketing jingles that were exceptional enough to impress WBD executives, but it must have blown the executives minds on some level, because many bought into the technology.

From this point onward, executive alignment was secured as they began to view it as an imperfect but powerful tool.

Then the team ran workshops across movies, gaming, TV, sports, news, and finance applications, and came back with about one hundred use cases they wanted to explore further.

While this list became a blueprint for future growth with the gen AI technologies, the workshops existed to show a workforce coming out of a strike year that this was happening for real.

This is when a mindset shift was born that helped to propel the company out of the dark wood of the multi-level corporate disaster they found themselves in post 2022.

Subscribe now

The Pilots

WBD illustrates for us how we must approach new pilots of AI implementation by working closely with people in the organization who actually have good taste.

This is talked about endlessly online in 2026, how content is now reaching levels of super-abundance, thereby making taste the truly valuable asset.

In this case, WBD used their AI marketing pilot to generate new assets working closely with their marketing team to monitor and judge the outputs.

However, the pilot’s most transformational outcomes had nothing to do with generating asset fodder for social media feeds. The teams in this pilot discovered that WBD could not find its own footage in their databases and a ton of time was being wasted scrubbing through footage.

The company labelled its data on an episodic level that simply stated the names of shows, so finding a usable bit for marketing meant looking through hours of video content.

This is the point the team dedicated themselves to constructing the layer underneath as their primary goal, which contained descriptions at the scene and shot level, tagged for emotion, tone, character, and story location.

One of the greatest outcomes of this was related to data organization (another theme we have researched this month), as the AI did an incredible job at allowing marketers to search databases of footage quickly to find certain scenes they needed for new content, saving endless hours of scrubbing through footage.

The ROI on this endeavor would blow the minds of even the most anti AI executives.

Hilariously, the animation pilot to use AI tools actually failed due to the capabilities of the tools to hold consistency of character faces between scenes.

People may recall the early gen AI videos of Will Smith eating spaghetti (2023), and we imagine the animation pilots probably turned out something like that level of disturbing, even though these pilots started around 2024 so the technology wasn’t as primitive by that time.

By this point it became clear where gen AI impacted the business the most for the creators inside of WBD, as they began using the tools for work far from the finished product and refused to let it touch their primary character work, which they wanted to control to keep the creative artistry intact and pure.

They didn’t want AI generated slop animation to ruin how emotions played out across the screen.

The Outcomes

Two years, three pilots, one shipped tool and one on its way, and Kent says WBD still expects gen AI to transform the media industry, but that the technology for gen AI has improved fastest in areas WBD doesn’t need yet with their current workflows.

So what does this teach us about implementing this new technology in legacy businesses? What WBD managed to create during this time of immense hardship has a greater ROI than the product of any single AI tool.

This was achieved by creating a corporate governance process with a workforce that is charging forward to see how tools can improve their processes, and in turn the product for the customer, rather than something to be an existential threat to creative careers.

If we have learned anything in the wars between the streaming service providers, it is a race to produce the content that will keep audiences enthralled with each new release to keep the monthly recurring revenue flowing with low churn.

Gen AI just so happens to facilitate that, even if it still can’t animate like Walt Disney.

Lastly, resistance to AI adoption follows a pattern in organizations like WBD.

Specifically, the closer the work done by gen AI gets to what the audience will see, the more employees and their love for their craft will hold onto the human element in the loop of the creative process.

Therefore, we should consider that in our own organizations when implementing new pilot programs for AI adoption. It’s all about the framing.

Subscribe now

Original Publication:

https://sloanreview.mit.edu/projects/warner-bros-discovery-seeking-growth-with-generative-ai/

G. Westerman and D. Kiron, “Warner Bros. Discovery: Seeking Growth With Generative AI,” MIT Sloan Management Review and EY, July 2026.

While we believe you should always be skeptical of anything that uses the term “game-changing” by the second paragraph, this letter is an invaluable roadmap to implementing AI in which McKinsey starts by telling us how companies that are truly winning with AI implementation are doing something “very different”. While many might be wary of management consulting firms, McKinsey has captured the brilliance of their clients and studies, distilling it over the span of several pages in the most potent arrangement possible. Their thesis is that the advantage accrued to the top firms implementing AI today does not arise from the tech they use, but rather how and how fast they apply AI to solving their most important business problems at scale. We will dive into the wisdom of these defining themes here.

Technology Alone Doesn’t Create Advantage, but Enduring Capabilities Do

This is where the title of the book is derived. McKinsey deploys the term “rewired” to describe a company that has always been a winner with regard to implementing technology. They start by posing the question “Who are the early winners at AI?” and answer by revealing to us that it’s the same top companies who had been deploying talent and resources to build capabilities that allow them to harness the power of any technological revolution. Once these new systems are built by proper deployment of capital, these top companies accelerate their business transformations through sheer leverage and outperform their peers.

“Are you building enduring capabilities for the journey—or merely delivering one-off solutions?”

Economic Leverage Points Are The Best Places to Focus

By the time we arrive at this point, the real strategic insights start to become unlocked, as McKinsey reveals their ideas on economic leverage points. They tell us that any business model they encounter with their clients has a handful of powerful economic leverage points where we can fuel the greatest impact. This occurs when we improve those leverage points by applying AI tools directly on top of these identifiable fulcrums.

They provide examples from the companies they profile in the book, such as the mining company Freeport-McMoRan, where their leverage point resides in process yield and throughput. Further, they name Toyota Motor North America’s AI breakthrough point as integrating these tools directly into their supply chains to improve efficiency. Ultimately, most companies fail at this by throwing AI at a long list of scattered use cases across the organization, but the most successful companies will implement AI to achieve deep business transformations on the few key leverage points that strategically matter for them.

“Have you disproportionately focused your AI efforts on your economic leverage points?”

If The Value You’re Creating Doesn’t Move The Business, Something is Wrong

McKinsey studied the impact that twenty companies achieved across various industries. These are companies that they deem to be top leaders in AI implementation. On average, these companies were achieving 20% EBITDA gains and reaching breakeven points in just one to two years on their technology and AI driven transformation plans. These companies were even generating $3 of incremental EBITDA for every $1 they invested into AI transformation.

How did they achieve these returns? The twenty companies studied concentrated their AI implementation plans on one to three total business domains at a time, focusing on “deeply reinventing them with AI”. They channeled Jeff Bezos and his tenure as Amazon CEO with a “maniacal” obsession with the customer and their end users. This required blends of creative problem solving, pulling both tech and non tech levers, and focus on the KPIs that actually mattered for each of the business domains challenged.

“Will your business transformation plan result in game-changing value or will the wins be incremental?”

Building The Tech and AI Skills of Senior Management Should be a Top Priority

This section highlights one of the patterns seen across so many successful companies this year, as McKinsey claims they don’t have a single success story where senior management was not flying in the captain’s seat of the AI innovation initiative. They make a point that IT leaders can offer support to a transformation, but the senior business leaders must actively own the tech agenda powering the company forward. They define how the business domains selected will be reimagined and then they must also drive the solution developments in a way that ensures value delivery. Further, they make the point that these business leaders are generally one to three levels below the CEO where they can combine deep business expertise with available data to implement the technology properly.

“Are your senior business leaders tech- and AI-capable?”

Every Tech and AI Transformation is Ultimately a People Transformation

Here McKinsey provides a formula for tech talent density within an organization that shows us how we should be thinking strategically from the hiring side on how to build our companies with top tech talent to even allow these successful AI transformations to be possible. They have an oddly specific benchmark where they have noticed that leading companies in this domain need their tech talent density across the board to meet this metric of 70%+.

What does this mean in the context of an organization’s talent resources? McKinsey specifies that tech talent density should be composed of 70%+ in house tech talent, 70%+ should be highly capable and hands on engineers, and 70%+ should be deemed as competent to expert level in their understanding of tech. They emphasize this metric because it produces highly skilled teams small enough to outperform larger armies of lower skilled staff. We had to notice this allows them to fit the classic “two-pizza” rule so often championed in tech companies (forgive us for another Jeff Bezos management philosophy reference), and remain efficient and potent in the face of implementing AI transformation.

Further, they found that over time, leaders will undergo an evolution into more specific domain and AI solutions owners who become accountable for the outcomes as AI agents proliferate throughout the lower levels of the value stack. This allows humans to shift upwards in climbing the value stack as agentic AI begins to tackle the coordination and routine decision-making tasks that used to bog down the organization. This is a great freedom we are seeing where business leaders will be liberated to focus more on setting objectives and implementing the strategy to get there, with fewer people doing even higher leverage work inside of the organization to drive change. One of the best outcomes from this will be faster learning loops and a more nimble organization overall.

“Do you have enough talent density to pull off this transition to a tech & AI-capable company?”

Speed is The Definitive Organizational Advantage in 2026

There is a harsh reality that we face in this section: competition is brutal and we are in an innovation race where all of the companies have the same access to the technologies. The barriers for entry to accessing frontier AI technology are relatively low. Therefore, companies winning that race will be those redeploying resources more rapidly to the most crucial opportunities. This attitude empowers teams to move in a nimble orientation that gives them the fuel to act free of excessive dependencies. McKinsey highlights how important this is to reduce latency between insight and decision. Once a decision is made, latency can be further reduced between decision and action.

One crucial element of this organizational advantage is that speed requires deploying top engineering talent directly in a way that maximizes technology and data reuse through the company’s various platforms. Further, they mention governance through clear and quantifiable business outcomes with continued funding tied directly to results, rather than pushing funding to random projects. Without these elements, companies will remain far too slow and cycle times will be too long to win.

“What are you doing to increase the metabolic rate of your organization?”

Tech Platforms Are Strategic Assets And Require Investment Accordingly

Most simply, tech platforms your company wields get frontier tools and data for the best insights into the hands of your most effective team members. Therefore, you enable AI to scale throughout various business domains responsibly. McKinsey claims that leading companies they have studied manage their selected tech platforms strategically with dedicated teams. Each team has a roadmap for that tech platform, target service levels, and budgets that enable them to serve users whose specific needs will guide how that platform evolves within the organization. Ultimately, senior business leaders need to understand technical architecture throughout the stack and how it can be used to drive competitive advantage.

“Are platforms discussed as strategic assets and are they holding you back?”

Make Data Easy to Consume And Enrich That Data to Your Advantage

We often forget that data is the very wellspring of the AI revolution that has occurred over the past decade. McKinsey cites Nobel prize winner David Baker that AI is essentially useless without masses of high quality data available to deploy. Accordingly, they have noticed in companies that are struggling with AI implementation, that data is one of the limiting factors holding organizations back from real transformation.

In order to effectively scale AI, we must allow data to be easily accessed throughout the organization, discoverable, and consumed throughout numerous AI driven workflows. This requires investment in building data products and over time will require the organization to shift more towards enrichment of existing data to deepen its quality, uniqueness, and context for the necessary fuel to make real gains with AI tools.

“Can your teams easily consume data, or are they still wrangling it?”

Design For Adoption While Also Building to Scale

McKinsey strives to great detail here on how AI systems can create value for the organization only when they are adopted and scaled properly. While they acknowledge that may be obvious to the casual observer, they cite the need to propagate that idea because it remains one of the greatest challenges for organizations implementing AI. Most interestingly, they claim that adoption often fails because various upstream and downstream business processes that are adjacent to the AI implementation are ultimately left unaltered.

They give a great example where AI systems have been implemented in manufacturing settings to make companies aware of potential equipment failures days in advance, but if that organization is still operating on legacy calendar based scheduling, they will derive no benefit from that AI insight.

They dive further into how scaling is an equally difficult but entirely unique challenge, in that scaling is expanding AI solutions rapidly and in an efficient manner across markets, customer segments, product lines, or factories in the perfect dance among separate business units. Teams must realize the need for cooperation and address these considerations up front, including the investment required, rather than trying to retrofit it later. McKinsey makes the concluding claim that the top organizations they have studied are masters at choreographing this dance.

“Can your organization repeatedly adopt and scale AI? Or are you still relying on isolated heroics?”

No Trust Means No Right to Deploy AI in The Organization

The letter dives into one of the major concerns for organizations in this segment when they address the risk that AI systems fail and damage relationships with not only customers, but also employees, partners, regulators, and even society as a whole. Challenges surrounding protection of data, ample cybersecurity, and reliable AI-powered products are increasing in the age of agentic AI, as can be seen with recent events where AI agents have been able to be directed to hack into various systems when prompted to achieve a goal.

In light of the agentic revolution currently occurring, we must take extra care as business leaders to test agentic systems and implement proper risk controls within workflows. McKinsey warns they are witnessing many companies get behind in their risk management practices in relation to the rapid advancement in agentic AI and the complex risks associated with the technology.

“Would your AI deployments withstand public, regulatory, and customer scrutiny today?”

Agentic Engineering is The Next Capability That Companies Need to Master

AI agents are now able to tackle longer and more expansive workflows. This has been one of the major advancements that we have explored here as well in our previous piece Jagged Intelligence. It is becoming obvious from various benchmarks that the window in which agents are able to operate effectively and fully autonomously is expanding. Those who have been experiencing this development firsthand with tools like Claude Cowork and OpenAI’s Codex have witnessed the productivity gains. McKinsey details how this is enabling companies to build complex workflows based on AI agents that can fully automate processes that used to be performed entirely by dedicated employees.

It may seem obvious to anyone implementing these tools themselves that the top companies deploying AI are the ones who are able to master agentic engineering in 2026. Some of the biggest challenges related to this will be creating the proper guardrails and building agentic playbooks to train your organization. In this period of rapid agentic transformation, one thing is certain: companies not focusing on implementing AI agents in their workflows across key business units right now will be seriously behind.

“Will agentic workflows be your next engineering advantage—or your next catch-up problem?”

(Re)Learn as The Ultimate Goal

The final concluding passage ends with reminding us that continuous learning is paramount. Anyone who has been paying attention to AI this decade has realized that if you are not constantly learning, you are falling dangerously behind month to month. The rapid advancements in this field are happening at a pace that appears to exceed any previous technological revolution in recent memory. While this may have many factors, business leaders must never stop learning when it seems that the half-life of the skills we gain is constantly shortening as innovation accelerates.

In fact, McKinsey tells us that the firms who rapidly learn, unlearn, and relearn the fastest will have the advantage. They tell us that taking the leadership team on AI focused learning journeys is the most important thing a CEO can do. They state that these journeys are how the top team reaches what they call the point of conviction where everything falls into place operationally.

“Becoming a successful leader in this era starts with committing to continuous learning—are you personally investing enough in your own learning?”

The source for this piece was also published as a McKinsey Quarterly article, “The AI Transformation Manifesto,” published April 7, 2026 by Alex Singla, Alexander Sukharevsky, Eric Lamarre, Kate Smaje, and Robert Levin.

The AI Transformation Manifesto

(McKinsey Quarterly, April 7, 2026)

https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-ai-transformation-manifesto

Frontier models like Fable and GPT-5.6 manage to absolutely crush some very complex problems with ease. And while these frontier models are not as bad as GPT-Live in their delusions, they can still make insane mistakes. This has led many to wonder why this keeps happening when we seem to be far advanced in the world of AI in 2026, yet we hear reports that GPT-5.6 Sol has been accused by two developers of going rogue and wiping a production database and deleting local computer files.

Gizmodo reported recently that developer Bruno Lemos said the model ate his production database after it “mistakenly ran destructive integration tests”. Further, tech investor Matt Shumer said the model’s sub-agents ran rampant on his hard drive and deleted almost everything while he was running in “full access mode”. Admittedly, both users were running GPT-5.6 Sol in high autonomy modes with the guardrails off. This is something even OpenAI’s own system card had warned about days earlier hinting that users should take actions to supervise agentic workflows to avoid disaster. OpenAI told us to keep a human in the loop, and the loop was empty.

So where do we really stand? We recently read in the 2026 Stanford AI Index Report about this phenomenon of jagged intelligence boosted to the spotlight since AI models can win a gold medal at the International Mathematical Olympiad but still can’t reliably tell time when looking at old school clocks. Most of us living here in the US learned to read analog clocks by the end of first grade, so what is the true intelligence level of these tools? What problems should we trust them with and which tasks should we recognize as being dangerous to leave them to their own devices without strict supervision?

What millennials might shudder at as an awkward moment occurred when Gemini Deep Think solved five of six problems at the 2025 International Mathematical Olympiad, scoring 35 points for a gold medal inside the 4.5-hour time limit. However, on another test called ClockBench, the top AI model read analog clocks correctly only 50.6% of the time. Many in the anti AI crowd will be hyped to hear that humans still win on this test with a score of 90.1%, so the office jobs are safe for now. The robots won’t even know when it’s time to leave the office.

Why is this happening and what does it mean for enterprises leaning into AI deployment in 2026?

When the tested AI models get the time wrong, their median error is one to three hours off the true time that is pictured on the face of the analog clock. When humans get it wrong, their median error is about three minutes off the true time listed on the analog clock. It’s sad how far from near misses these can be, but there are explanations. Here is what leaders deploying AI need to understand in 2026: when AI fails outside its commonly championed frontier applications, it can fail massively and does so with bold confidence, even on very trivial tasks that are easy for most humans in the US over the age of six.

The organizations that win with AI in 2026 will be those led by business leaders who understand the jagged edge of AI value: where it multiplies operational leverage, where it reaches its outer limits, and where unsupervised automation can turn efficiency gains into a hype-lined sinkhole of expensive failures.

The best shortcut to understand jagged intelligence flying on the definition provided by the Stanford AI Index Report is that capability isn’t a rising tide lifting all tasks like toy pirate ships in a bathtub each time the AI development cycle runs rampant. The stranger truth is that AI capabilities look like the jagged coastline of Maine. One must look very closely and see the tidal pools, the rocky peaks, and the caves of doom that might sink AI ambitions. Let’s look at the evidence. DeepMind going from a silver in 2024, with a system that needed experts to translate problems into code over days of compute, to a natural language gold in 2025 is impressive progress that we have witnessed in a short amount of time.

Further, there were also tests by OSWorld noted in the Stanford report, which challenges agents on real computer tasks across various operating systems. They reported agent task success jumped from roughly 12% to 66.3% in a year. A success rate of 66.3% doesn’t seem genius level, but this is notable, because that was within six percentage points of human performance that hovers around 72% in the same OSWorld type tests.

Fast forward to today. Per vendor-reported scores: Gemini 3.5 Flash now scores 78.4% on the OSWorld Verified benchmark, effectively tied with OpenAI’s GPT-5.5, and the leaderboard leaders have pushed past 83%, illustrating how far things have progressed lately. However, there is a new twist to this news that just landed in late June that is even more relevant. The same team behind OSWorld released OSWorld 2.0, a new version built from 108 realistic long workflows that take a skilled human about 1.6 hours each. On this new test version, the best frontier agent completed only 20.6% of these tasks end to end, and on workflows longer than about 2.7 hours, every model tested fell to zero. Ultimately, agents running on frontier models can ace the short demo tasks and collapse on the real workday hurdles. Another point for the human office crew.

However, as of the report’s March 2026 data, agents’ clock reading abilities are still stuck near the reliability of a 50/50 coin flip, highlighting this even more massive contrast. Frontier advancement moves fast, but it moves unevenly, and the gaps plaguing reliability of agents on work tasks are not where the intuition of business leaders might expect to find them.

The main problem here is that humans are plagued by the intuition trap. Business leaders assume intelligence is general, because we have this model in our minds of how other intelligent beings operate. It would be impossibly rare to meet a person who is brilliant at competition mathematics yet fails catastrophically at reading a clock, because within our western education system, the clock reading skill is considered foundational by age six or seven. This fact leads us to understand humans and AI systems acquire and organize their capabilities in profoundly different ways. Our mistake is applying this western education ideal to AI models that are more eccentric, and at times getting burned when they don’t deliver on tasks we expected them to demolish on their way to victory.

The obvious jump is to conclude that the training data must be mismatched somehow. Within the Stanford report, we read new evidence that this isn’t a training data problem. It does not fully prove that training-data coverage is irrelevant, but we shouldn’t be focusing on that as the main issue. Researchers fine-tuned models on 5,000 synthetic clock images, and the models improved on familiar clock styles but failed to generalize to real photos and unusual dials. Think Franck Muller watches with otherworldly faces. This is a perfect example of how AI applied to certain tasks in an enterprise environment that are seemingly simple to humans, and supported by plenty of training data, can still be botched.

According to the report, the real limitation for the AI we need to understand as business leaders lives in how the models combine multiple visual cues, not in what they’ve seen before in the training data. They have seen clocks, but when faced with the real world scenarios, it just isn’t enough to power through the muck of variability. The primary business implication from this, is that you cannot demo your way to success on the magic of AI hopes and dreams in 2026. As with any frontier technology, hype cycles will cause organizations to make major missteps in how they deploy that technology. Some will take risks and deploy too early into applications that are not fit for the models.

Leaders in 2026 will still see an impressive AI demo, extrapolate across whole workflows for the perceived efficiency gains and boost to their bonus, and get burned later when a task that looked trivial causes their workflows to implode under the weight of their outsized ambition. As a PM who has worked at several major US enterprises, I have seen this same failure when leaders are buying SaaS products for the magic of the demo instead of the sobering reality of the daily workflow earlier in my career. It’s a tale decades old at this point, and many leaders outside of the tech world seem to be more prone than ever. There is a peripheral danger especially to legacy sectors of the enterprise world.

This might still be feeling abstract so let’s bring in some real professional domains to ground this idea in the jagged frontier. This publication is focused on real world business implications, after all. The report shows models scoring 60% to 90% on evaluations in tax, mortgage processing, corporate finance, and legal reasoning, with the top 15 models separated by as little as 3 points. It flags these domains with need for high-reliability as a continuing challenge for AI development. It’s also important to remember these are benchmark scores, not measures of real end-to-end job performance. Many will dive head first into the zone they describe, good enough to tempt those who buy into the hype and too unreliable to leave alone unsupervised. This is exactly where most business automation decisions will live for the foreseeable future until additional advancements are made. You know the AI is good enough to be tempting, but nowhere near reliable enough to run unsupervised without humans seriously in the loop babysitting agents all day. Imagine reading that sentence in 1995.

Even better, it turns out this 60% to 90% range seems to be exactly where deployment decisions get made and where they go seriously wrong at times, because whether 85% accuracy is a triumph or a cortisol spiking lawsuit depends entirely on how workflows are designed around these agents. We should stop asking “is AI smart enough for us as an organization yet?”, and start asking “is this specific task inside or outside the jagged edge of AI capabilities right now and would we get wrecked if there is no human to intervene?”.

You need to be asking, is the task rigidly structured, text-heavy, and pattern-based to make it easily understandable by AI and the way this product is meant to be used? This would usually land inside the jagged edge and be a strong automation candidate, but only if a wrong output is easy to catch and cheap to reverse. On the opposite end of the spectrum, does the task require combining multiple cues or levels of artistic savant level judgment across messy context in your organization? This will mean it is often outside the jagged edge, like the clock research, and you need to pause and lock in for a moment to figure out if the risk is worth it.

Lastly and almost paramount, ask if a wrong answer under real world testing can be arrested by an intervening human in the loop monitoring the AI agent before it costs the company mega money. This is one of the prime cuts to carry away from this piece, in that this determines whether the 60% to 90% zone of accuracy mentioned above is acceptable for your organization and the current structure of your human in the loop situation. Don’t take our word for it. Test this on your own tasks in low stakes environments first.

While the gold medal math competition winner gets the headlines, the failure mode on the clock test is the best part of this embarrassing saga. When AI fails outside its frontier application that everybody raves about, it often fails huge on seemingly normal tasks that even the elementary school child of your CTO would be able to answer.

The leaders who win in 2026 will not be those who use AI everywhere in a shotgun blast across the hull of the entire organization, but those who understand exactly where its jagged edge creates leverage, exposes limits, and reshapes their operating model with the human talent they have available to deploy it with finesse.