With the long holiday weekend in the US, I want to leave you with a lot to consider.
From Fast Company, some guidance on AI use policies. Research indicates that in organizations with a substantial workforce, such as a tech company with tens of thousands of engineers, the adoption rate of AI tools fell below 50%, primarily because employees associated their use with a lack of skill, despite identical output quality. This mindset is prevalent across various businesses, leading to unproductive policies that misunderstand the true value of AI. Instead of focusing on disclosure requirements, companies are encouraged to shift towards an ownership model where employees take full responsibility for their outputs, regardless of the tools used. This approach promotes accountability and encourages the effective use of AI as just another business tool, rather than a controversial entity. By changing the conversation from “Did you use AI?” to “Is this excellent?” organizations can fully leverage AI’s potential to improve business results.
Are we in a bubble? The Washington Post has a good read which I’ve linked to, as according to JPMorgan, the tech industry must generate an additional $650 billion in annual revenue to justify these expenditures through 2030. In contrast, OpenAI claims it will soon reach a revenue rate of $20 billion annually. But Crazy Stupid Tech has a comparison to the internet bubble of 1999. It highlights that AI has rapidly evolved into a global phenomenon within just three years, raising concerns about potential overspending and inflated valuations. For instance, global AI capital expenditures and venture capital investments have already surpassed $600 billion in 2025, with projections suggesting that total spending could reach $1.5 trillion by year-end, according to Gartner. Major tech companies are heavily investing, yet uncertainty remains about whether these expenditures will yield sustainable profits, as exemplified by OpenAI’s substantial cash burn despite its high valuation. Analysts warn of the risks associated with excessive leverage and the concentration of financial exposure among a few key players, raising concerns about a potential market correction.
The New Stack looks at How Autonomous Agents are Changing Infrastructure Management, with the average cost of downtime now estimated at $12,900 per minute, and nearly $24,000 for large enterprises. These AI DevOps engineers function as autonomous agents that analyze infrastructure, coordinate operational tools, and propose actions in near-real time. Unlike traditional automation, they integrate directly with enterprise cloud environments and support existing governance frameworks. Many organizations now require that infrastructure-related data remain within their cloud accounts, particularly those in sensitive sectors like healthcare and finance, which has shifted the focus towards cloud-native large language models such as Amazon Bedrock. This trend reflects a growing commitment to improving operational efficiency while ensuring compliance and security in infrastructure management.
And I’ll leave you with Amazon’s Chief Technology Officer, Werner Vogels, predictions for 2026, emphasizing the rising role of artificial intelligence and its implications for both developers and society. In a recent conversation, Vogels indicated that while he will not be retiring, this may be his final keynote at the annual AWS re:Invent event, which he has attended for over two decades. Vogels’ predictions focus on five key areas, including the use of technology to combat loneliness, the evolution of developers into “Renaissance” thinkers, and the urgent need for quantum-safe encryption due to advancing data harvesting techniques. He noted that many military innovations have historically transitioned into civilian applications, yet emphasized the ethical considerations surrounding such technologies. Additionally, he highlighted the importance of personalized learning to foster curiosity in children, arguing that nurturing this trait is essential for effective education.
Why do we care?
Here’s the thing I want you thinking about over the long weekend: we are treating AI like it’s both magical and fragile, and neither mindset is helping. If your team feels embarrassed to use AI because they think it signals they’re not good enough, your AI policy is already broken. You want people accountable for the quality of work — not whether they typed it themselves. That’s how you get real adoption. Your AI policy has to show up in your contracts — MSAs, SOWs, DPAs, and SLAs need to spell out who owns AI decisions, who audits them, and what happens when they fail.
But there’s a second thread here: the money side of AI looks increasingly disconnected from the reality of AI deployment. If JPMorgan thinks the industry needs another $650 billion of revenue to justify current spend, and the biggest player is only talking about hitting $20 billion, that should make you pause. A bubble doesn’t kill innovation — it kills companies, and those companies are often the ones you’re depending on for your stack. MSPs depend on VC-backed AI vendors that may not survive a correction, which could break customer workflows overnight.
Then add autonomous agents. These things will be managing infrastructure before most businesses are ready to govern them. Great for efficiency, terrible if you don’t know what the agent is doing behind the scenes. And the vendors will pitch this aggressively. MSPs will soon have to offer AI policy, oversight, and auditing as services—not just deployment and support. You’ll need logging, telemetry, and review loops for AI decisions the same way you do for backups and security events.
And finally, Vogels reminds us that AI isn’t just an IT story anymore. It’s a social story. A policy story. A workforce story. That means the choices you make today will matter a lot more than you realize.
Which leads to the questions you should be asking:
- Do my current AI policies encourage or discourage people from actually using AI?
- Which vendors in my stack are most exposed if the AI spending bubble corrects?
- How much of my roadmap depends on companies that may not survive a consolidation wave?
- What guardrails do I need before an AI system starts making infrastructure decisions?
- If AI fails silently in my workflows, how quickly would I know?
- Who owns the responsibility when AI-generated work is wrong — my staff, the vendor, or me?
- Am I building an AI practice around tools, or around outcomes?
- How do I communicate AI risk and value to customers without feeding into the hype?