The AI Tool Adoption Curve: Where We Are Now

Technology adoption follows curves. We have known this since Everett Rogers described it in 1962 and Geoffrey Moore refined it for technology in 1991. The S-curve, the chasm, the crossing from early adopters to mainstream — these frameworks have held up remarkably well across technology cycles.

AI is following these curves, but in a more compressed and layered way than previous technology transitions. Understanding where different segments of AI adoption sit on the curve — and why — tells you what is happening, what is coming, and where to position yourself.

The Three Waves#

AI adoption is not one curve. It is at least three overlapping waves, each on its own adoption timeline.

Wave 1: AI features in existing products. This is the most mature wave. Copilot in Office, Einstein in Salesforce, AI in Gmail, suggestions in Grammarly. Most enterprise knowledge workers have been touched by this wave. Adoption is wide but depth is shallow — people use these features occasionally, for specific tasks, without fundamentally changing their workflows.

Wave 2: AI-native tools. ChatGPT, Claude, Perplexity, Midjourney, and similar. These are standalone AI products used by a smaller but highly engaged audience. Power users have built real workflows around them. Most mainstream users have tried them but not integrated them deeply. This wave is in the mainstream chasm — broad awareness, patchy adoption, variable depth.

Wave 3: Agent-native systems. Full agent stacks where AI does real work autonomously — like DenchClaw, Devin for code, autonomous research agents. This wave is early adopter / innovator territory. Most organizations are watching but not using. The chasm is still ahead.

The interesting thing about these waves: they interact. The successful crossing of Wave 2 accelerates Wave 3 adoption. The organizations that build deep Wave 2 workflows develop the operational muscle for Wave 3.

Wave 1: The Broad but Shallow Wave#

Wave 1 AI features are nearly universal in enterprise software now. Almost every major platform has shipped AI features in the last 18 months.

The adoption pattern here is: feature launches, gets initial press, gets used by the most curious users, settles into occasional use by a subset of the user base. Deep integration is rare.

The reason shallow adoption persists: Wave 1 features are designed to fit into existing workflows, which means they are limited by the constraints of those workflows. Copilot in Word is constrained by the document. Einstein in Salesforce is constrained by the Salesforce interaction model. The AI suggests; the human still does.

There is still adoption growth in Wave 1 — companies are still rolling out these features to employees who have not yet used them. But the productivity floor here is relatively limited. These features make existing workflows modestly faster. They do not change the fundamental structure of work.

Wave 2: The Fragmented Middle#

Wave 2 is where the interesting action is right now. ChatGPT, Claude, Perplexity, and a dozen other AI-native tools have achieved significant reach among knowledge workers.

But the adoption is fragmented in a specific way: there is a group of highly engaged power users who have built genuinely transformative workflows around these tools, and a much larger group of occasional users who try them for specific tasks but have not integrated them deeply.

The chasm here is the context gap. Power users have invested time in building context into their workflows — writing detailed system prompts, creating custom GPTs or projects, maintaining prompt libraries, building document stores the AI can reference. This investment produces dramatically better outputs.

Occasional users prompt from scratch each time. They get decent but generic outputs. They never experience the productivity leap that comes from a contextually rich AI.

Crossing the Wave 2 chasm requires helping users bridge the context gap. Products that make it easy to build and maintain context — to give the AI what it needs to produce specific, useful outputs — will pull mainstream users across.

Wave 3: The Early Agent Frontier#

Wave 3 is where agent-native systems live. This is early-adopter territory.

The organizations experimenting with Wave 3 are primarily: AI-native startups building their own tools, larger technology companies with dedicated AI teams, and individual operators who have been in Wave 2 long enough to see the ceiling.

The chasm ahead is trust. Wave 3 tools require giving an agent real authority to take real actions. Most organizations are not ready to do this, because they do not yet have frameworks for understanding what the agent is doing, catching mistakes, and maintaining accountability.

This is not irrationality. It is appropriate caution. The tools and frameworks for responsible agent deployment — constraint design, audit logs, reversibility, escalation protocols — are still being built. Wave 3 mainstream adoption will not happen until these tools are robust and well-understood.

What Is Crossing the Chasm Right Now#

Here is my read on where different AI capabilities sit in the adoption curve:

Fully mainstream: AI-assisted writing (Copilot, Gmail Smart Compose), AI search (Perplexity, Bing AI), AI code suggestions (Copilot, Codeium). These are used without second thought by millions of professionals.

Early mainstream: AI-generated first drafts for standard documents (proposals, reports, emails), AI-summarized meeting transcripts, AI-powered research workflows. Enthusiast adoption is high; mainstream adoption is accelerating.

Late early adopter: Fully AI-operated customer support tiers, AI-generated marketing content pipelines, AI agents with real write access to operational systems. Teams are experimenting; few have gone fully mainstream internal.

Early adopter: Full agent stacks like DenchClaw, autonomous research agents, AI with real browser automation and action-taking capabilities. The edge, not the mainstream.

Innovator: Fully autonomous AI operations (agents operating without scheduled human review), AI-AI agent collaboration, agent swarms. Still experimental and high-risk.

What This Tells Us About Opportunities#

Each zone on this curve has a different opportunity profile.

The fully mainstream zone is competed. Building a better AI writing assistant is a crowded market. The commoditization is happening now.

The early mainstream zone has volume opportunity. Massive markets are crossing right now. Products that make AI-generated documents and meeting intelligence robust and trustworthy at scale will capture a lot of value.

The late early adopter zone has strategic opportunity. The teams and products that cross this chasm well will have significant head starts when the mainstream follows. This is where I think the best ROI on investment is right now for serious AI practitioners.

The early adopter zone has pioneer opportunity. Risky, but the organizations that figure out trustworthy agent deployment now will have a 2-3 year head start. That is the bet we made with DenchClaw.

The Acceleration Effect#

One thing that makes this technology cycle different from previous ones: the curves are compressing.

The PC adoption curve played out over a decade. The web over a decade. Mobile over a decade. AI features went from announcement to broad enterprise adoption in 18 months. AI-native tools went from GPT-3 to mainstream in roughly 2 years. The timeline is 3-5x compressed compared to previous technology cycles.

This matters for strategy. The window between early adopter and mainstream is narrower. The head start you get from being early is measured in months, not years. The urgency to invest in building context, designing workflows, and developing organizational muscle for AI is correspondingly higher.

Where Are You on the Curve?#

If you are a founder or operator, the honest question is: which wave are you on?

Using AI features in existing tools (Wave 1) but not doing more: you are behind the productivity curve and the gap is widening.

Using AI-native tools regularly but without deep context integration (Wave 2 occasional): you are using these tools at a fraction of their potential. The investment in context and workflow design will produce disproportionate returns.

Building agent-native workflows with real tool access and persistent context (Wave 3 early adopter): you are developing the operational muscle that will be a genuine competitive advantage as the mainstream crosses this chasm.

The goal is not to be on any particular wave. It is to be intentional about where you are and what it would take to move forward on the curve.

Frequently Asked Questions#

How long before AI agents are mainstream in enterprise?#

My estimate: 2-3 years for early mainstream (forward-thinking enterprises), 4-5 years for broad mainstream (crossing the chasm). The pace depends heavily on how quickly trust tools and frameworks mature, and how quickly the early adopter cohort produces case studies that reduce perceived risk.

Are we in an AI bubble, or is this different from previous technology overhype cycles?#

The hype is real and some of it will deflate. But the underlying productivity gains from current AI capabilities are also real — more real than the productivity claims of most previous technology cycles. The tools that are genuinely useful will survive the correction. The ones built on hype alone will not.

What's the biggest risk for organizations adopting AI agents too early?#

Deploying agents without adequate constraint and oversight infrastructure. Early-mover advantage is real, but early-mover mistakes can be expensive and erode organizational confidence in AI in ways that set back adoption for years.

How does DenchClaw fit into the adoption curve?#

DenchClaw is a Wave 3 product — agent-native, with real tool access and persistent memory. It is designed for early adopters who are ready to go beyond AI features and AI tools to genuine agent operations. As Wave 3 crosses mainstream, the organizations that have been running DenchClaw will have accumulated context advantages that are difficult to replicate quickly.

Ready to try DenchClaw? Install in one command: npx denchclaw. Full setup guide →