Epilogue 2035: After the AI Bubble

A field report from the future on what remained once the AI hype cycle broke.

I learned my first real lesson about technology leadership during the dot-com bust. I was an early-career programmer at Texas A&M, working on what then felt like cutting-edge projects: moving core business processes onto the web. Around us, the economy was faltering. State funding dropped. Hiring freezes spread. Anxiety moved quickly through the IT organization. In the middle of that uncertainty, our director, Tom Putnam, did something simple and rare. He stood in front of the staff, laid out the numbers line by line, and explained exactly how the organization would proceed. No spin. Just clarity. The work did not stop. In fact, some of the most durable systems of that era were built precisely because the hype had faded and discipline returned.

That experience shapes how I view the current AI boom. The technology is real. Much of the hype is not. In today’s Dispatch, I want to apply the lessons of the Internet’s boom and bust to the present moment by projecting forward to 2035 and asking a quieter, more durable question: what remains of AI after its boom and bust cycle ends. The aim is not to slow adoption, but to help leaders adopt AI now in ways that will still hold when capital tightens, expectations reset, and the real work begins.

The big picture

From the vantage point of 2035, the AI boom of the mid-2020s no longer feels novel or unpredictable. Capital was abundant, and anxiety was widespread. Every CEO wanted an AI strategy, quickly. Leaders spoke in absolutes. Vendors promised transformation at scale. Institutions moved fast, often faster than their data, culture, and governance could support. Speed became proof of seriousness. Caution was framed as resistance.

This was not new.

The dot-com era followed the same arc. Vision surged ahead of capacity. Charisma outran discipline. The technology did not fail. The economics did. When capital tightened, the gap between promise and readiness became impossible to ignore. AI followed that same path. The boom ended not in collapse, but in clarification.

The thesis is straightforward. The bust did not kill AI. It revealed what it was actually for. Real innovation did not emerge during the period of abundance. It emerged after it. By 2035, AI is no longer discussed as a breakthrough. It is infrastructure.

The noise faded. What remained was useful.

What the Boom Looked Like While We Were in It

The atmosphere of the mid-2020s was defined less by the technology itself than by the social pressure surrounding it. Organizations stood up AI transformation offices. Job descriptions shifted faster than job content. Employees worried quietly about replacement while leaders felt pressure to move before fully understanding what they were buying. Acting quickly became a proxy for seriousness. Caution was reframed as resistance. The only legitimate question was whether you would lead or fall behind.

That pressure had a structural cause. Capital was abundant, and abundance removes the friction that disciplines decision-making. When money is available and the competition is moving, the cost of waiting feels higher than the cost of a mistake. Vendors understood this and priced their proposals accordingly. Contracts grew large and long-term. GPU costs were treated as inconveniences rather than constraints. Custom models were launched without a clear path to maintenance or return.

The data wall arrived quietly. Institutions discovered that generic models built hastily on top of large language models were too prone to error and too disconnected from core institutional data to be trusted for real decisions. The unified, governed data required to make AI agents reliable and predictable simply did not exist in most organizations. Pilot projects proliferated but were rarely evaluated with rigor. Productivity claims multiplied faster than evidence. The gap between demonstration and production-grade performance was wider than the rhetoric had suggested.

What changed was not belief in the technology. It was the recognition that the conditions enabling rapid adoption, abundant capital, diffuse accountability, and deferred governance, had masked that gap between hype and reality rather than closed it. When those conditions shifted, the gap became visible and expensive.

What Actually Survived

Once capital tightened and enthusiasm cooled, a smaller, sturdier set of patterns remained. These are the changes that defined AI’s long-term impact.

AI Became a Base Layer, Not a Headline

By 2035, drafting, summarizing, and first-pass analysis, the early cognitive work of surfacing patterns and framing problems, are no longer treated as skills. They are utilities, as ordinary as spellcheck or search. AI did not replace most professionals. It compressed layers of low-complexity work. Tasks that once absorbed hours became instantaneous. One person, properly supported, could now do what several once did.

The constraint shifted. Judgment became the differentiator. The people who advanced were not the best prompt engineers, but those who could frame problems clearly, assess outputs critically, and accept responsibility for decisions. That shift exposed a simple truth: the use of AI amplifies judgment, it does not replace it. AI reshaped how work was done, but it did not remove the need for thoughtfulness and accountability.

Standalone Tools Disappeared into Systems

The boom years were defined by portals. Log in here. Ask the model there. Leaders told staff to “use AI” without changing how work actually flowed. The assumption was that exposure alone would increase efficiency. It did not. These tools lived at the edges of work rather than inside it, disconnected from the systems where decisions were made and value was created. Adoption stalled as complexity eclipsed curiosity.

By the early 2030s, AI had moved into the core. It was embedded inside ERP, CRM, finance, HR, and operational platforms, handling reconciliation, anomaly detection, routing, and forecasting quietly. We stopped treating these systems as software and started treating them as “digital workers,” entities with defined roles, access privileges, and performance metrics, formally integrated into the org chart alongside their human counterparts. No one talked about “using AI” anymore. They talked about closing the month faster, reducing errors, and handling more volume with the same staff. AI became plumbing. That shift mattered because it aligned technology with measurable outcomes instead of novelty, and once aligned, it finally endured.

The cycles of 'log in to this AI portal' faded. The lasting change wasn't just embedded AI inside software, but the rise of autonomous agents working across that software. By 2035, you don't ask a chatbot to summarize a customer issue. An agent notices the issue, crosses silos to investigate shipping and billing data, processes the return, and updates the CRM; only alerting a human if it hits a roadblock. AI stopped being a dutiful assistant waiting for a prompt and became an active participant in operations.

Governance Stopped being Aspirational

During the boom, AI governance was largely rhetorical. Organizations published principles and ethical statements, but enforcement was thin, and controls were immature. Responsibility was diffused. Risk was treated as theoretical. That posture did not survive sustained use, public scrutiny, or the first real failures.

By 2035, model risk is managed the way cybersecurity risk was a decade earlier. Data lineage is tracked. Outputs are audited. Boundaries are explicit. Some decisions remain automated. Others are deliberately human. This shift was not driven by virtue. It was driven by lawsuits and liability. Institutions relearned a familiar lesson from the Internet era. Trust does not emerge from intention. It requires structure and process.

At the same time, human accountability became a much more critical resource. Demand surged for those with the discernment to review outputs for nuance, the courage to make strategic decisions under uncertainty, and the emotional intelligence to negotiate complex human dynamics. The primary function of the knowledge worker shifted profoundly from being the engine of execution to becoming the supervisor of autonomous agents. Their value is defined by their ability to manage the critical exceptions where models failed, to rigorously define the ethical and operational guardrails before the work began, and to step in specifically for those high-stakes, sensitive interactions that still demanded genuine human empathy.

Implications for AI-Infused Universities

In higher education, the period after the boom clarified that becoming an AI-infused university was not about strategy or choosing a tool. It was about aligning different forms of AI to different kinds of work. The institutions that adapted best understood that not all AI was the same, and not all use cases carried the same risk, cost, or payoff. Over time, four distinct domains emerged, each requiring a different posture.

1. Research and discovery: game-changing AI, inward-focused. This was where large-scale, high-cost AI investments worked. High-performance computing and specialized models accelerated research in areas where scale and complexity mattered. The payoff was not automation, but amplification. Literature review, data exploration, and hypothesis testing compressed dramatically, allowing researchers to spend more time on theory, design, and interpretation. The universities that succeeded treated this infrastructure as shared and mission-critical, not as a collection of bespoke labs. Governance mattered here, but so did patience. These investments paid off over time, not in headlines.

2. Administrative operations, process, and ERP: game-changing AI, customer-focused. This domain delivered some of the most visible gains for students, faculty, and staff. Embedded inside ERP, CRM, advising, finance, and HR systems, AI reduced friction that had long defined the university experience. Routing improved. Reconciliation became faster and more accurate. Forecasting became more reliable. Importantly, these gains came from deep integration into core systems. Universities that treated AI as part of process redesign, rather than an overlay, finally saw durable improvements in service and efficiency.

3. Personal productivity for employees: everyday AI, inward-focused. Most gains came not from licensing tools, but from disciplined use of AI already embedded in productivity platforms. Drafting, summarizing, analysis, and preparation became faster and more consistent. The real investment was in training. Those institutions that focused on AI literacy helped employees move from execution to orchestration, enabling more critical and skeptical thinking rather than replacing it. Everyday AI raised the cognitive bar for staff work rather than lowering it.

4. Teaching and learning: everyday AI, customer-focused. In the classroom, AI did not replace teaching. It reshaped expectations. As first-pass work became automated, assignments shifted toward interpretation, synthesis, and ethical reasoning. Faculty emphasized how to question AI output rather than how to produce it. Students learned that AI was a tool to engage with, not a shortcut to avoid thinking. Universities that succeeded framed AI literacy as a core educational outcome, not a technical skill, preparing students for a world where judgment, context, and responsibility mattered more than speed.

Taken together, these four domains explain why the AI-infused university was quieter and more disciplined than many expected. The most successful institutions did not deploy AI everywhere at once. They matched the right kind of AI to the right kind of work, invested deliberately, and governed consistently. Overall, they built capabilities that survived the hype cycle and remained valuable long after the boom ended.

The Leadership Lesson that Endured

The boom rewarded a particular kind of leader. Vision was currency. The ability to tell a compelling story about the future and move quickly in its direction signaled serious leadership. Skepticism was reframed as resistance, or the sign of someone who’s behind. That phase of the hype cycle feels decisive while it lasts, and it always ends.

When capital tightened, the center of gravity shifted. Vision gave way to stewardship. Leaders were no longer judged by the ambition of their promises, but by what they could sustain under pressure. Institutions confronted the risks and tradeoffs they had minimized previously. Scarcity imposed the discipline that abundance had deferred.

The technology market followed this correction. The bust pushed vendors away from bloated per-seat licensing toward consumption models where businesses paid only for work completed by autonomous agents. Smaller, more efficient models displaced the assumption that scale alone produced value. Institutions that invested in data quality and process clarity were positioned to take advantage of those shifts. Those who had not found themselves holding expensive infrastructure with uncertain returns.

What the correction clarified, more than anything, was the relationship between credibility and outcomes. The leaders who lasted were not the ones who had been most enthusiastic. They were the ones who consistently managed the space between promise, resources, and results. They resisted innovation theater and focused on measurable efficiency, reduced risk, and clear ownership when those qualities were less celebrated. By 2035, those qualities were the ones that defined success with AI.

The final word

From the vantage point of 2035, the AI boom looks less like a turning point and more like a rite of passage. Every major technology follows this arc. Exuberance gives way to overreach. Overreach invites correction. Maturity comes later. The Internet followed this path. AI was never going to be different. The mistake was not enthusiasm. The mistake was believing enthusiasm could substitute for discipline.

By 2035, AI is no longer exciting. It is reliable. It is embedded. It is constrained.

The institutions that emerged stronger were not the ones that moved fastest. They were the ones who prepared for the moment when budgets tightened and questions sharpened. They trained people to orchestrate, not just execute, recognizing that AI required deeper thinking, not less of it. They invested in boring systems that scaled and worked under pressure, the core systems through which work is done in 2035.

From the other side of the hype cycle, the lesson is clear. Do not fear the bust. That is where the real work begins.