The BeAIReady Brief | Week 20
May 11–17 | Most AI Strategies Are 'More for Show' Than Substance, the Workforce Is Restructuring Faster Than Leaders Are Managing It, and the Human Infrastructure Required to Close Either Gap Doesn't
Economists have been cautious about interpreting the April job numbers as “stability”: 115,000 jobs added, unemployment holding at 4.3%, a labor market that looks, from a distance, like it’s finding its footing. But the information sector — tech, telecom, data processing — logged its sixteenth consecutive month of net job losses, even as the four largest tech companies committed roughly $725 billion to AI infrastructure this year.
That gap between AI investment and AI employment is starting to wear thin among organizational leadership. The consistent finding is that while AI works — organizations aren’t. The tools remain ahead of the institutions using them, and that widening gap is starting to show up in the workforce, in the ROI numbers, and in the psychological cost to the people caught in the middle.
Here’s what I’ve been reading.
The ROI Isn’t Missing — the Org Is
Multiple independent data sources converged on the same uncomfortable finding last week: most organizations are failing to turn AI deployment into business value, and the reasons have nothing to do with the technology.
AI Restructures the Workforce — and Organizations Are Letting It Happen
From Meta’s 8,000 layoffs to collapsing graduate hiring to the death of equal raises, last week’s reading showed a workforce already sorting itself in ways most organizations haven’t consciously designed for.
The Human Infrastructure Nobody Built
Three pieces converged on a single problem: the psychological readiness, skills development, and team collaboration infrastructure that would make AI adoption actually work — that’s the part nobody bought a license for.
What Microsoft Is Shipping
New data from the Work Trend Index, a meaningful governance update to Copilot Studio, and a question about whether the Chief AI Officer role signals organizational maturity or something more transitional.
On the Bigger Picture
One engineer’s argument that the app interface itself is dying, and one startup’s claim that turn-based AI conversation already belongs to the past.
The ROI Isn’t Missing — the Org Is
The most striking number I came across last week was buried in Writer’s 2026 enterprise AI adoption survey of 2,400 executives and employees: 97% of organizations have deployed AI agents in the past year, and only 29% see significant ROI from generative AI. Three-quarters of executives admitted their company’s AI strategy is “more for show” than actual internal guidance. 48% percent called AI adoption a “massive disappointment.” The gap between deployment and transformation is widening, and the data makes it clear this is a governance and culture failure — not a capability failure. (Enterprise AI adoption in 2026: Why 79% face challenges despite high investment)
Microsoft’s 2026 Work Trend Index, analyzed in a Fortune piece aimed at CFOs, arrived at a compatible finding from a different angle. Organizational factors — culture, manager support, talent practices — account for 67% of reported AI impact, compared with just 32% attributed to individual mindset and behavior. If two-thirds of AI’s business value depends on how an organization is structured and led, then framing AI ROI as a technology problem doesn’t just misdiagnose the issue — it gives the wrong people cover for the gap. The alignment numbers underscore it: only 26% of AI users say their leadership is clearly and consistently aligned on AI strategy, and only 13% say they’re rewarded for reinventing work with AI even when results aren’t immediate. (What Microsoft’s new research tells CFOs about the ROI of AI)
A CIO.com piece introduced a concept I’ll be using for a while: the “aversion tax.” The argument is that every dollar of AI investment is subject to a reduction based on the actual adoption rate — an algorithm that is 99% accurate but only 10% utilized has effectively destroyed 90% of its value. The author cites an Amazon case where a model identifying $176 million in annualized savings sat unused because floor managers reverted to gut instinct and sticky notes rather than trust the machine. The implication for enterprise leaders is that any honest ROI calculation for an AI initiative has to start with a realistic human adoption estimate, not a feature spec — and most of them don’t. (The ghost in the machine: Why AI ROI dies at the human finish line)
Goldman Sachs’ CIO Marco Argenti offered the most useful reframe on measurement I read all week. He doesn’t track how much each employee uses AI — even though Goldman can see that data for all 12,000 of his engineers. He tracks how fast a team moves from idea to prototype. “There’s zero time between idea and prototype — you kind of ‘3D print’ software,” he told Business Insider. The right metric isn’t usage frequency, it’s delivery velocity: if your backlog isn’t shrinking, the AI investment isn’t landing, regardless of what the adoption dashboard shows. That shift in measurement philosophy — from individual tool engagement to team-level output — is one of the cleaner frameworks I came across this week. (Goldman Sachs’ tech boss says he doesn’t track AI usage, he watches how fast teams ship)
AI Is Restructuring the Workforce — and Organizations Are Letting It Happen
The April 2026 jobs report data showed that the information sector — tech, telecom, data processing — shed another 13,000 jobs. That’s makes sixteen consecutive months of net losses, bringing payrolls to their lowest level since March 2021, wiping out four years of sector gains. This is happening simultaneously with the largest AI infrastructure spending in history. The sector that is supposed to be the primary beneficiary of AI investment is losing jobs at a rate that is one of the longest peacetime declines in any major sector in modern labor data — and the causal link to AI is still being debated precisely because that debate is uncomfortable. (April 2026 jobs report: AI, white-collar layoffs, wages)
Meta made the structural logic explicit. With 8,000 layoffs scheduled for May 20 and capital expenditure climbing to a record $125–145 billion this year, Zuckerberg described a company being rebuilt around what he calls “ultraflat” teams — one manager for every 50 engineers — where AI tools allow one or two people to ship in a week what once required dozens over months. What’s striking isn’t the headcount number, it’s that CFO Susan Li admitted she doesn’t know what the company’s ideal size even looks like anymore when AI capabilities keep shifting what one person can build — and that honesty should unsettle every executive making workforce planning assumptions based on last year’s productivity baselines. (Meta to cut 8,000 jobs on May 20)
The graduate hiring data makes the structural shift tangible at the entry level. Adzuna reported a 34.9% decline in graduate vacancies in the year to March, even as overall job postings rose. Economists remain cautious about attribution, but the direction is consistent: entry-level white-collar roles — the ones where organizations build institutional knowledge and develop the next generation of managers — are taking the first clear structural hit, and the long-term implications for organizational depth and talent pipelines haven’t been seriously reckoned with yet. (Graduate jobs fall by a third as more employers embrace AI)
The compensation story running alongside all of this is equally consequential. A Mercer report found that only about 4% of U.S. employers actually gave equal “peanut butter” raises this year, despite earlier surveys suggesting 44% were considering it. The reason is becoming clearer: AI super-users are outperforming peers at rates that make uniform pay feel inequitable to anyone delivering more. Google has begun incorporating AI usage into software engineer performance reviews, Accenture has made AI fluency a requirement for promotion, and the Writer survey found that AI super-users were three times more likely to have received a raise or promotion in the past year — which means the performance management system is already sorting people by AI capability whether organizations have designed it to or not. (Companies ditch ‘peanut butter’ raises as pay-for-performance takes over)
The Human Infrastructure Nobody Built
An HBR piece introduced the concept of “psychological debt” in AI adoption — six distinct forms of psychological cost that unstructured AI use creates for employees:
Cognitive - skills atrophying through offloading
Autonomy - loss of control over how work happens
Competency - feeling less capable relative to the machine
Relatedness - diminishing peer connection
Credibility - perceived loss of professional standing from using AI
Identity - AI use conflicting with professional self-concept
The survey data accompanying the framework was pointed: employees who use AI rarely reported psychological debt scores of 60, versus 36 for daily users — meaning the people who most need to adopt it are accumulating the most friction against doing so, which is the opposite dynamic of what most organizations are planning for. (The Psychological Costs of Adopting AI)
The skills story runs parallel. DataCamp’s 2026 survey of 500+ enterprise leaders found that 82% offer some form of AI training and 68% say employees have access to learning resources — but only 35% report a mature, organization-wide AI upskilling program. Among those that do, reports of significant AI ROI nearly double, from 21% to 42%. The gap isn’t investment in training — it’s that most AI training is passive, generic, and disconnected from actual workflows, which produces awareness without capability; and organizations keep measuring training completion rates rather than the behavioral change that determines whether any of it lands. (AI Skills Gap in 2026: Why Training Isn’t Enough)
A five-month HBR experiment with 60 managers added the team layer. When groups tried to use AI collaboratively in meetings without any structured approach, the AI defaulted to responding to whoever was typing, not to the group. Teams fell into passive spectator mode within the first session, and the AI effectively narrowed participation rather than widening it. Three deliberate practices reversed this: introducing the team to the AI collectively, assigning the AI multiple rotating roles (challenger, customer, skeptic), and maintaining shared ownership of every prompt rather than letting one person drive. Average team engagement increased 30% after those practices were in place, and two-thirds of participants said their group alignment had improved — which suggests that the ability to use AI productively in team settings is a distinct organizational capability that doesn’t emerge from individual AI literacy alone. (It’s Hard to Use AI as a Team. These 3 Practices Can Help.)
What Microsoft Is Shipping
(Disclosure: my company, StitchDX, is a Microsoft partner. The coverage below reflects my read of publicly available announcements.)
An IBM study published last week found that 76% of the more than 2,000 organizations surveyed have now established a Chief AI Officer role, up from 26% in 2025. The CNBC piece covering it raised the more interesting question: is the CAIO a permanent C-suite fixture, or a transitional designation created to navigate a specific moment of AI integration that will eventually be absorbed into the CIO, CHRO, or COO? The answer matters because it determines how organizations scope and staff the role — and a transitional designation built for short-term navigation is a different job than a permanent seat responsible for compounding organizational AI capability over time. McKinsey’s framing in the piece — that coordinating AI across a company is more important than any specific title — is probably the more durable principle. (Do you need a chief AI officer? Here’s how the tech is changing boardrooms)
The Microsoft 365 Copilot blog published the companion post to the Work Trend Index, and the concept I found most useful was what it calls the “Transformation Paradox”: 65% of AI users fear falling behind if they don’t adapt, but 45% say it feels safer to focus on current goals than to redesign work with AI. What Microsoft is naming — and what the WTI data underscores — is that the same urgency driving AI adoption is also producing the organizational caution that prevents it from compounding. Resolving that paradox is a leadership task, not a tooling one. The post also announced that Copilot Cowork is now available on iOS and Android, and that the first wave of federated Copilot connectors — including HubSpot, Moody’s, and Notion — are generally available in Microsoft 365. (Microsoft 365 Copilot, human agency, and the opportunity for every organization)
The Copilot Studio April update addressed the governance problem directly. The new Analytics Viewer role separates performance visibility from configuration rights — stakeholders can see how agents are performing without the ability to modify them. The expanded agent usage estimator now forecasts Copilot credit consumption across both Copilot Studio and Dynamics 365 environments, shifting budget management from guesswork to data. Agent 365 is now generally available as the centralized control plane for managing agents across the full Microsoft environment — the governance layer that organizations running agentic workflows at scale have been waiting for. GPT-5.5 Reasoning is also now available in Copilot Studio early-release environments. (What’s new in Copilot Studio: April 2026 updates and features)
On the Bigger Picture
A former Google principal engineer who spent eight years building Google Sheets wrote one of the more clarifying pieces I read last week. His team recently built a working Sheets clone in a few days — not as good as the real thing, but closing fast. His argument: when building an app takes days instead of years, the app itself is worth less. Value is moving away from the interface and toward the data underneath it — and companies that have built defensibility around a front-end experience sitting on top of a database are closing a window faster than most of the people inside those companies want to believe. What replaces it, in his framing, is the “meta-app”: AI tools that generate custom applications on demand, where the user describes intent and the system figures out the rest. The SaaS implications are significant. (I spent 8 years building Google Sheets. Now I think apps are on their way out)
Thinking Machines — Mira Murati’s post-OpenAI startup — previewed a genuinely different class of AI interaction model. Rather than the standard turn-based exchange, their TML-Interaction-Small uses full-duplex architecture that processes 200-millisecond chunks of input and output simultaneously — listening, talking, and seeing at the same time. The turn-taking latency is 0.40 seconds, compared to 1.18 seconds for GPT-realtime. The enterprise implications of sub-second, continuously aware AI — real-time safety monitoring, live translation that feels like conversation, time-aware process management — are significant, but the model is still in limited research preview and hasn’t been tested at production scale. Worth tracking; not yet worth building plans around. (Thinking Machines shows off near-realtime AI voice and video conversation with new ‘interaction models’)
The bottom-line
Organizations are restructuring around AI faster than they’re building the capacity to use it well. That’s showing up in how workforces are being sorted — by compensation, by hiring, by headcount — and along lines drawn in the past eighteen months.
The problem is, the human infrastructure that would make that restructuring work — the psychological readiness, the skills that transfer, the team norms, the governance, the incentive alignment — is lagging… badly. In some organizations, it hasn’t even started.
The uncomfortable implication is that most of the reorganization currently underway is happening in the absence of the thing it requires to succeed. That’s an urgent organizational leadership problem that AI is putting into sharper focus.
That’s it for this week’s BeAIReady brief!
If you appreciate the depth of reporting and how I connect the dots, please like, share this post, and subscribe (or share the Brief with a friend!). Thanks!
~erick