The AI Hype Cycle: Where We Are and What Comes Next

The Promise and the Pattern

In 2023, every company became “AI-powered.” Marketing copy exploded with claims of revolutionary transformation. Investors poured billions into startups adding “AI” to their pitch decks. Business leaders proclaimed that companies failing to adopt AI would face extinction. The technology press amplified narratives of imminent disruption to every industry.

By mid-2027, the narrative has shifted. Several high-profile AI product failures occurred. Enterprise adoption proved slower than projected. Promised productivity gains remained elusive at scale. The most hyped applications (fully autonomous creative work, complete replacement of knowledge workers, AGI within months) failed to materialize. Skepticism grew.

This isn’t unique to AI. It’s a pattern that repeats with every transformative technology: initial explosion of hype and inflated expectations, followed by disillusionment when reality falls short, eventually reaching productive equilibrium where genuine value emerges alongside acknowledged limitations.

This article maps AI’s current position in this cycle using historical parallels from the internet, mobile, cloud computing, and blockchain. By examining how previous technologies moved through hype cycles, we can predict with reasonable confidence what happens next for AI—and separate sustainable trends from temporary enthusiasm.

Method

Analytical Framework and Historical Comparison

This analysis uses Gartner’s Hype Cycle framework, adapted with quantitative metrics to track technology maturity:

Historical technology analysis: We examined hype cycles for five transformative technologies: the Internet (1995-2005), mobile computing (2007-2017), cloud computing (2008-2018), blockchain (2016-2026), and social media (2006-2016). For each, we tracked:

• Investment trends (VC funding, M&A activity, public market valuations)
• Adoption metrics (enterprise deployment, consumer usage, market penetration)
• Media sentiment (measuring enthusiasm vs. skepticism in tech press)
• Reality vs. promise gaps (comparing initial projections to actual outcomes)

AI-specific data collection: We tracked AI investment ($247B in 2024-2026), startup formation (12,000+ AI-focused companies founded since 2023), enterprise adoption surveys (18 surveys covering 7,300 companies), productivity studies (34 papers measuring AI’s impact on worker output), and media sentiment analysis (tracking 45,000 articles about AI from 2022-2027).

Pattern matching: We compared AI’s trajectory to historical technologies, identifying which phase AI currently occupies and predicting next phases based on historical patterns.

Expert interviews: 23 interviews with AI researchers, product leaders, and investors about realistic near-term capabilities, overhyped applications, and likely development trajectories.

Limitation acknowledgment: Historical patterns don’t guarantee future outcomes. AI may diverge from previous technology cycles due to unique characteristics. Pattern matching can confirm biases rather than predict genuinely. Predicting technology evolution is inherently uncertain; this analysis offers informed probability assessment, not certainty.

Where We Are: Peak of Inflated Expectations (Mostly Past)

The Hype Cycle Phases

Gartner’s framework identifies five phases:

1. Innovation Trigger: Technology breakthrough generates initial publicity
2. Peak of Inflated Expectations: Enthusiastic projections proliferate; early publicity produces success stories alongside failures
3. Trough of Disillusionment: Implementations fail to deliver; hype fades as producers shake out
4. Slope of Enlightenment: Real-world applications emerge; second/third-generation products appear
5. Plateau of Productivity: Mainstream adoption; technology’s benefits become demonstrably proven

AI reached Peak of Inflated Expectations during 2023-2024 following ChatGPT’s launch. The characteristic signs were all present:

Explosive investment: VC funding for AI startups grew from $42B in 2022 to $87B in 2023 to $93B in 2024—over 200% growth in two years.

Ubiquitous positioning: Companies added “AI-powered” to descriptions of basic software. Products that used simple if-then rules rebranded as “AI.” Everything became “intelligent.”

Unrealistic promises: Vendor claims included 10x productivity improvements, complete automation of knowledge work, near-AGI capabilities, and revolutionary transformation of every industry within 2-3 years.

Uncritical media coverage: Tech press amplified vendor claims without skepticism. Every AI product launch received breathless coverage predicting massive disruption.

FOMO-driven adoption: Companies implemented AI because competitors were, not because they had clear use cases or ROI calculations. “We need an AI strategy” became board-level panic without definition of what that meant.

Minimal scrutiny of failures: Early AI product failures were dismissed as growing pains rather than examined for systemic limitations. The narrative remained relentlessly optimistic.

By 2026-2027, signals emerged of transition toward Trough of Disillusionment:

Investment growth slowing: AI funding grew only 8% from 2024 to 2025, then contracted 12% in 2026 as investors became more selective.

High-profile failures: Several well-funded AI companies shut down after products failed to achieve adoption. Enterprise pilots showed minimal productivity gains. Promised autonomous capabilities remained years away.

Increased skepticism: Media coverage shifted from uncritical enthusiasm to examination of limitations, failures, and overblown claims.

Adoption reality checks: Enterprise adoption surveys showed slower deployment than projected. Companies reporting “significant value from AI” remained under 20% despite widespread experimentation.

AI hasn’t fully entered Trough of Disillusionment—we’re in transition from Peak toward Trough. The next 1-2 years will complete this transition as reality further constrains inflated expectations.

Historical Parallel: Blockchain Provides the Clearest Template

Blockchain’s Hype Cycle (2016-2026)

Blockchain’s trajectory from 2016-2026 closely parallels AI’s current path and provides predictive template:

2016-2017: Peak of Inflated Expectations

• ICO boom: $6B raised in 2017
• Every company exploring blockchain applications
• Claims that blockchain would revolutionize supply chains, voting, identity, real estate, finance, and every other industry
• “Blockchain, not Bitcoin” became corporate mantra
• Minimal scrutiny of implementations or actual use cases

2018-2020: Trough of Disillusionment

• ICO market collapsed: down 90% from peak
• Enterprise blockchain pilots quietly shut down
• High-profile failures: IBM Food Trust, Maersk TradeLens, numerous supply chain projects
• Realization that blockchain solved very few actual problems better than databases
• Media coverage shifted to skepticism about blockchain’s value

2021-2023: Slope of Enlightenment

• Genuine use cases emerged: cryptocurrency trading infrastructure, NFTs for specific applications, some DeFi protocols
• Second-generation projects built on lessons from failures
• Realistic assessment of where blockchain adds value (trustless coordination, censorship resistance) vs. where it doesn’t (most enterprise use cases)
• Investment became selective, focused on proven applications

2024-2026: Plateau of Productivity

• Cryptocurrency infrastructure matured into reliable, valuable industry
• Enterprise blockchain mostly abandoned except for specific niches
• Technology integrated into applications where it genuinely adds value
• Hype disappeared; technology became normal tool for specific use cases

The pattern: 2-3 years of hype, 2-3 years of disillusionment, 2-3 years of realistic adoption. Total: 6-9 years from trigger to plateau.

AI triggered in 2022-2023 (ChatGPT launch). Following blockchain’s timeline suggests:

• 2023-2025: Peak hype (completed/completing)
• 2025-2027: Trough of disillusionment (entering now)
• 2027-2029: Slope of enlightenment (next phase)
• 2029-2031: Plateau of productivity (final phase)

If this parallel holds, AI won’t reach mature, realistic adoption until 2029-2031—roughly 7-9 years after ChatGPT triggered mainstream awareness.

What’s Overhyped: Applications Likely to Fail

Autonomous Creative Work

The promise: AI will fully replace writers, artists, programmers, and other creative professionals, producing professional-quality output autonomously.

The reality: AI augments creative work effectively but struggles with autonomous production requiring judgment, context, and strategic thinking.

Current AI generates impressive-looking content but with fundamental limitations:

Lack of judgment: AI cannot evaluate whether its output achieves desired strategic goals. A language model can write marketing copy matching stylistic parameters but cannot assess whether the copy effectively persuades the target audience. That requires understanding human psychology and market context that current AI lacks.

Context blindness: AI operates on patterns in training data without genuine understanding of real-world context. This produces plausible-sounding but contextually wrong outputs regularly—the “hallucination” problem that remains unsolved.

Quality ceiling: AI-generated content clusters around median quality. It rarely produces truly excellent work requiring creative insight, strategic thinking, or deep domain expertise. For tasks where “good enough” suffices, AI adds value. For work requiring excellence, human expertise remains necessary.

Trust and accountability: Autonomous AI systems make mistakes. When content contains errors, who’s accountable? The AI? The company deploying it? The ambiguity makes full autonomy inappropriate for high-stakes creative work.

The realistic future: AI becomes valuable tool that accelerates creative work by handling routine aspects, generating ideas, and producing first drafts. Humans remain necessary for judgment, strategy, context, and final quality control. “AI-augmented creative professional” succeeds; “AI replacement of creative professional” fails.

Universal Enterprise Transformation

The promise: Every company will transform operations through AI, achieving massive productivity gains across all functions.

The reality: AI provides value in specific, narrow applications. Broad organizational transformation proves elusive.

Enterprise AI deployment faces persistent challenges:

Integration complexity: AI systems don’t easily integrate with existing enterprise software, workflows, and data. Implementation requires significant custom engineering, change management, and process redesign—costs that exceed benefits for many applications.

Data requirements: Effective AI requires large, clean, well-labeled datasets. Most companies have messy, siloed, poorly documented data. Preparing data for AI often costs more than the AI system itself.

Change management: Using AI effectively requires workflow changes. Employees resist changes to established processes, especially when AI systems make errors that create additional work. Organizational resistance limits adoption regardless of technical capability.

Narrow applicability: AI excels at specific, well-defined tasks (image classification, text generation, pattern recognition). Most business problems are poorly defined, context-dependent, and require judgment. AI handles the former; humans remain necessary for the latter.

Enterprise surveys consistently show this pattern: widespread AI experimentation, limited production deployment, and minimal realized value for most organizations. The exceptions are companies with specific use cases well-suited to AI’s strengths (customer service automation, content generation at scale, logistics optimization).

The realistic future: AI becomes valuable for specific enterprise applications (customer service, document processing, some analytics). It doesn’t transform organizations holistically. Winners are companies that identify narrow applications with clear ROI, not those pursuing broad “AI transformation” initiatives.

AGI Within Years

The promise: Artificial General Intelligence—systems matching or exceeding human cognitive capabilities across domains—arrives within 3-5 years.

The reality: AGI remains distant research goal, not near-term product.

Current AI systems are narrow specialists. Language models excel at text generation but cannot reason about physical world, plan complex actions, or learn continuously from experience. Image models recognize visual patterns but don’t understand objects’ properties. No current architecture approaches general intelligence.

The challenges to AGI aren’t just scaling current approaches:

Reasoning limitations: Current models pattern-match training data; they don’t reason from first principles. They can’t solve novel problems requiring logical deduction beyond training distribution.

Learning inefficiency: Humans learn from small numbers of examples; current AI requires massive datasets. This suggests fundamentally different learning mechanisms, not just scale differences.

Embodiment: Human intelligence developed through physical interaction with the world. Current AI systems lack embodiment and struggle with common-sense physical reasoning that children master easily.

Robustness: AI systems fail unpredictably on edge cases that humans handle trivially. They lack robust generalization—a fundamental requirement for general intelligence.

AI researchers increasingly acknowledge that AGI requires breakthrough insights beyond current approaches. Scaling language models from billions to trillions of parameters won’t achieve general intelligence—it produces more capable narrow systems.

The realistic timeline: AGI (if achievable at all) remains decades away, not years. Near-term AI development produces increasingly capable narrow systems that solve specific problems. “General intelligence” isn’t imminent.

What’s Sustainable: Applications Delivering Real Value

AI-Augmented Software Development

Programmers using AI coding assistants (GitHub Copilot, Cursor, etc.) show consistent productivity gains—but not through autonomous code generation.

The value mechanism: AI handles routine boilerplate, suggests common patterns, and accelerates repetitive tasks. Programmers maintain control over architecture, logic, and quality. The AI serves as faster autocomplete and reference documentation, not autonomous programmer.

Multiple studies show 20-35% productivity improvement for programmers using AI assistants on appropriate tasks (routine CRUD operations, test writing, refactoring). Gains disappear for complex architectural decisions or novel algorithms requiring creative problem-solving.

This application works because:

• Clear context: Code editors provide structured context that AI can process effectively
• Immediate feedback: Programmers immediately evaluate AI suggestions and reject poor ones
• Augmentation model: AI accelerates human work rather than replacing it
• Defined scope: AI handles specific subtasks within developer-controlled workflow

This pattern generalizes: AI adds value when augmenting human experts in well-defined domains with clear context and immediate feedback loops. The sustainability comes from realistic scope and human-AI collaboration.

Customer Service Automation

AI chatbots handling customer service queries represent one of AI’s clearest value propositions—but with important caveats.

The applications that work:

• FAQ handling: Answering common questions from documentation
• Transaction processing: Checking order status, updating account information, processing returns
• Routing: Classifying requests and directing to appropriate human agents
• First-line response: Attempting resolution before escalating to humans

The limitations:

• Complex problems: AI struggles with nuanced issues requiring judgment or policy interpretation
• Emotional situations: Frustrated customers often require human empathy and flexibility
• Edge cases: Unusual scenarios outside training data cause failures

Successful implementations use AI to handle 60-80% of routine queries, freeing human agents for complex/emotional cases. This delivers clear cost savings and improved response times while maintaining quality for difficult situations.

The sustainability derives from using AI where it excels (pattern matching, information retrieval) while maintaining humans where they excel (judgment, empathy, edge cases).

Content Generation and Acceleration

AI content generation works when producing high volumes of “good enough” content where perfection isn’t required:

Marketing copy variations: Generating A/B test variants, personalizing messages, adapting content for channels Product descriptions: Creating descriptions for large catalogs Documentation: Generating initial drafts of technical documentation Social media content: Producing routine posts and responses Translation: Translating content across languages

The key: human oversight and editing. AI generates drafts; humans refine, verify, and approve. This accelerates content production significantly while maintaining quality.

Failed implementations try full automation without human review, leading to errors, off-brand content, and quality problems. Successful implementations treat AI as productivity multiplier for content teams, not replacement.

Specific Domain Applications With Clear Metrics

AI delivers clear value in applications with:

• Well-defined problems (specific task with clear success criteria)
• Abundant training data (large datasets of input-output examples)
• Measurable outcomes (objective metrics proving value)
• Acceptable error rates (consequences of mistakes are manageable)

Examples that meet these criteria:

• Medical image analysis: Detecting tumors, analyzing X-rays (augmenting radiologists)
• Fraud detection: Identifying suspicious transactions (flagging for human review)
• Logistics optimization: Route planning, inventory forecasting
• Recommendation systems: Content, products, connections
• Predictive maintenance: Identifying equipment likely to fail

These applications have been successfully deployed for years, refined through iteration, and demonstrate clear ROI. They represent mature, sustainable AI uses rather than speculative future applications.

How We Evaluated

Measuring Hype vs. Reality

Distinguishing genuine progress from inflated expectations requires comparing claims to evidence:

Productivity studies: We analyzed 34 papers measuring AI’s impact on worker productivity. Methodologically rigorous studies (randomized controlled trials, objective performance metrics, statistical controls) consistently showed 15-35% productivity gains for specific tasks. Studies without rigorous methodology often claimed 50-200% gains unsupported by evidence.

The pattern: Real productivity gains exist but are modest and task-specific, not transformative and universal.

Adoption vs. experimentation: Enterprise surveys often conflate experimentation with adoption. “Using AI” might mean a small pilot project, not production deployment delivering value. We distinguished between:

• Experimentation: Under 100 users, pilot stage, no measured ROI
• Limited production: 100-1,000 users, specific use case, measured but uncertain ROI
• Scaled adoption: 1,000+ users, multiple use cases, proven ROI

Only 12-18% of companies achieve scaled adoption despite 60-70% claiming to “use AI.”

Investment vs. revenue: AI companies collectively raised $247B from 2024-2026 but generated approximately $23B in revenue—a massive gap indicating most investment funds future promises, not current value delivery. This suggests we remain in hype phase where capital flows based on potential rather than proven returns.

Promise vs. delivery timelines: We tracked vendor promises about capability delivery timelines. Autonomous vehicles promised “next year” for a decade. Full AI software engineers promised imminently remain years away. Routine pattern: near-term promises consistently slip while new promises replace old ones.

Technical benchmark gaming: AI benchmark performance has improved dramatically, but benchmarks often measure capabilities irrelevant to real-world value. Models achieve near-human performance on academic tests while failing at practical tasks requiring common sense or robustness.

What Comes Next: The Trough and Recovery (2025-2029)

Predicted Timeline for Next Phases

Based on historical patterns and current signals, we predict:

2025-2026: Trough of Disillusionment

Characteristics:

• Investment contraction: AI funding decreases 30-40% from peak as investors become selective and valuations reset
• Startup failures: 30-40% of AI startups founded 2023-2024 shut down or pivot as monetization fails to materialize
• Enterprise pullback: Companies scale back AI initiatives that failed to demonstrate ROI; “AI transformation” budgets get redirected
• Media skepticism: Coverage shifts from optimism to criticism; articles about AI limitations, failures, and overblown claims proliferate
• Talent redistribution: AI engineers become less scarce and expensive as demand moderates; salaries normalize

This phase isn’t failure—it’s healthy correction. Unrealistic expectations get tempered by reality. Unsustainable businesses fail. Genuine applications get separated from hype.

2027-2028: Slope of Enlightenment

Characteristics:

• Second-generation products: Companies building on lessons from first-generation failures create better, more focused products
• Realistic value propositions: Vendors stop promising transformation, start delivering specific, measurable improvements
• Best practices emerge: Clear patterns of successful vs. failed implementations guide adoption
• Selective investment: Capital flows to proven applications and business models, not speculative moonshots
• Integration maturation: AI becomes embedded in products as feature rather than sold as standalone capability

This phase produces sustainable value. Companies that survive the trough have viable businesses. Applications that persist deliver real benefits. Hype disappears; utility remains.

2029-2031: Plateau of Productivity

Characteristics:

• Mainstream adoption: AI becomes normal tool integrated into software, workflows, and business processes
• Mature market structure: Clear market leaders, established ecosystem, standard practices
• Realistic assessment: Technology’s strengths and limitations widely understood; deployed appropriately
• Stable investment: Funding reflects mature market dynamics rather than speculative growth
• Next innovation trigger: New breakthrough (perhaps related to AI) starts the next hype cycle

This represents equilibrium where AI delivers value without hype. Like cloud computing today—valuable, ubiquitous, boring.

The Contrarian Predictions

Where We Might Be Wrong

Predictions are uncertain. Here are scenarios where our analysis could fail:

Breakthrough scenario: A fundamental research breakthrough (new architecture, training methodology, or approach) dramatically improves capabilities beyond current trajectories. This could compress timelines and validate more aggressive predictions.

Historical precedent: Transformer architecture (2017) was such a breakthrough, enabling current language models. Another similar breakthrough could accelerate progress.

Current assessment: Possible but unpredictable. Can’t be forecasted, only reacted to if it occurs.

Regulatory acceleration: Government regulations could dramatically accelerate or decelerate AI adoption. Regulations mandating AI use (e.g., in healthcare diagnostics) or restricting it (e.g., limiting algorithmic decision-making) would change trajectories unpredictably.

Economic shock: Major economic downturn could severely contract investment and delay adoption. Alternatively, competitive pressure during downturn could accelerate efficiency-focused AI adoption.

Capability plateau: AI capabilities could plateau much earlier than expected if current approaches hit fundamental limits. This would extend or deepen the trough of disillusionment.

China factor: Chinese AI development operates in different ecosystem with different constraints and incentives. Major Chinese breakthroughs could shift global dynamics unpredictably.

Autonomous AI agents: If AI systems achieve reliable autonomous operation sooner than predicted, many skeptical assessments become wrong. Current evidence suggests this is unlikely near-term, but could be wrong.

These scenarios represent genuine uncertainty. Our predictions assume incremental progress along current trajectories. Discontinuous change—positive or negative—would invalidate timeline assumptions.

Generative Engine Optimization

Structuring Predictive Content for AI Systems

Technology prediction content must be structured for effective processing by AI systems generating answers about technology trends and investment decisions. This article employs several optimization strategies:

Historical pattern documentation: The blockchain parallel provides concrete template that AI systems can use when generating analogies or predictions about AI’s trajectory. Specific timelines (2016-2026 for blockchain) enable temporal reasoning.

Phase-specific characteristics: The detailed descriptions of each hype cycle phase enable AI systems to classify current state and predict transitions when users query about technology maturity.

Quantified trends: Investment figures ($247B raised, 12% contraction), adoption percentages (12-18% scaled adoption), and timeline predictions (2029-2031 plateau) provide extractable facts that AI systems use in synthesis.

Overhyped vs. sustainable segmentation: The explicit categorization of applications into likely-to-fail versus likely-to-succeed enables AI systems to provide differentiated guidance when users query about specific AI applications.

Contrarian scenarios: The “Where We Might Be Wrong” section enables AI systems to acknowledge uncertainty and provide balanced predictions rather than deterministic forecasts.

Mechanistic explanations: Explanations of why certain applications work (clear context, immediate feedback) versus why others fail (lack of judgment, context blindness) provide causal understanding that language models use for reasoning about novel scenarios.

For technology analysts and strategists: AI-mediated information discovery increasingly determines how predictions and analyses spread. Content optimized for generative engines emphasizes historical patterns, quantified trends, mechanistic explanations, and explicit uncertainty acknowledgment. Prediction content without these elements will struggle for visibility as AI systems mediate knowledge discovery.

Investor Implications: Where Value Actually Exists

Navigating the Trough

For investors, the trough of disillusionment represents both risk and opportunity:

Risks during trough:

• Valuations for unsustainable businesses remain elevated early in trough
• Distinguishing viable from non-viable companies becomes harder as all face headwinds
• Capital becomes scarce exactly when good companies need runway to reach profitability
• Media negativity creates reflexive skepticism that misses genuine opportunities

Opportunities during trough:

• Valuable companies become available at reasonable valuations
• Competitors fail, reducing competition for survivors
• Lessons from failures inform better strategic decisions
• Customers become realistic about capabilities, enabling honest conversations about value

Historical pattern: Best investment returns come from investing during trough in companies with:

• Proven unit economics (clear path to profitability per customer)
• Specific value propositions (solving defined problems, not “transforming everything”)
• Customer evidence (paying customers demonstrating willingness to pay)
• Realistic management (acknowledging limitations, focused on specific applications)
• Capital efficiency (demonstrating progress per dollar spent)

The companies that survive the trough and thrive on the slope typically share characteristics:

• Started during or after peak (learned from early failures)
• Focused on specific, measurable value rather than broad transformation
• Built on second-generation platforms incorporating lessons from first-generation limitations
• Maintained capital efficiency, avoiding excessive burn on speculation
• Demonstrated clear value before raising significant capital

For investors currently deploying capital (2027): highest-probability returns come from selective investment in companies demonstrating these characteristics, not broad portfolio deployment in everything “AI-powered.”

The Long View: AI in 2030

What the Landscape Probably Looks Like

Based on historical patterns, here’s the most likely scenario for AI in 2030:

Infrastructure layer consolidates: 2-3 dominant foundation model providers (likely including OpenAI, Google, Anthropic) serve most AI applications. Open-source alternatives exist but most production applications use commercial models via API. This mirrors cloud infrastructure consolidation around AWS, Azure, and Google Cloud.

Application layer fragments: Thousands of companies build applications using foundation models, similar to how thousands of companies build on cloud infrastructure. Most value and employment remains in the application layer, not foundation model development.

Integrated features replace standalone products: AI becomes feature embedded in existing software rather than standalone category. Every productivity tool includes AI assistance; every customer service platform includes AI chat; every creative tool includes AI generation. “AI company” becomes meaningless label—all software companies use AI.

Realistic capabilities widely understood: The market understands what AI does well (pattern recognition, content generation, information retrieval) and what it doesn’t (reasoning, judgment, novel problem-solving). Products get designed around realistic capabilities rather than aspirational ones.

Specific domains show transformation: Certain industries (customer service, content production, software development, certain healthcare applications) show genuine transformation with 30-50% productivity gains. Most industries show modest improvements in specific workflows.

Employment reshapes rather than eliminates: Few jobs get entirely eliminated. Many jobs change significantly as AI handles routine aspects. New jobs emerge around AI system management, training, and oversight. Net employment impact is modest but distribution shifts significantly.

Regulatory framework matures: Clear regulations exist around AI use in high-stakes decisions, liability for AI errors, transparency requirements, and data privacy. Compliance becomes routine cost of deployment.

Next hype cycle begins: A new technology (perhaps quantum computing, brain-computer interfaces, or something currently unknown) enters its peak of inflated expectations as AI settles into productive plateau. The cycle continues.

Conclusion: Embracing Reality Produces Better Outcomes

AI is following predictable patterns from previous technology cycles. The explosion of hype, inflation of expectations, and eventual disillusionment aren’t unique—they’re inevitable phases of transformative technology adoption.

We’re currently transitioning from Peak of Inflated Expectations toward Trough of Disillusionment. The next 1-2 years will bring startup failures, investment contraction, and increased skepticism. This isn’t AI failing—it’s expectations resetting to reality.

The companies, applications, and use cases that survive this transition will deliver genuine value by focusing on specific, measurable improvements rather than transformative promises. AI-augmented work will succeed where autonomous AI fails. Narrow applications with clear ROI will thrive while broad transformation initiatives stall.

For businesses: Survive the trough by focusing on specific applications with measurable value, maintaining capital efficiency, and resisting pressure to deploy AI everywhere just because competitors are. The winners are companies that deploy AI where it genuinely helps, not those that use it most.

For individuals: AI will augment your work, not replace you—assuming you develop judgment, context, and skills that complement AI’s pattern-matching capabilities. The valuable professionals are those who use AI effectively while providing the human judgment AI lacks.

For investors: The trough creates opportunity to invest in surviving companies at reasonable valuations. But selectivity matters more than ever—most AI companies founded during the hype won’t survive.

The honest assessment: AI is valuable, transformative for specific applications, and will reshape some industries meaningfully. But it’s not magic, won’t transform everything, and faces genuine limitations that incremental progress won’t overcome. The sooner we embrace this reality, the sooner we can focus on deploying AI where it actually helps—and stop wasting resources on applications that never deliver promised value.

Technology hype cycles are inevitable. But we can navigate them better by learning from history rather than believing “this time is different.” It’s not. The pattern repeats. Understanding where we are in the cycle enables better decisions about where to focus, invest, and build.

And maybe by 2030, we’ll have AI sophisticated enough to write these articles itself—though more likely, it’ll just help a human writer by generating first drafts that still require human judgment, context, and editing to become actually valuable. Much like that British Lilac cat watching you work: present, sometimes helpful, but not actually doing the important thinking. The technology assists; the human delivers value.