Every technology wave arrives with prophecy. This one comes with demos.

Scroll through your feed and you will see stories of agents booking holidays, drafting legal briefs and replacing entire departments. Venture capitalists talk about trillion dollar shifts, while founders speculate about the end of work. Yet those deploying AI say they are working harder than ever[1].

AI will change the economy in profound ways. That does not mean you should build your strategy around what it might do in 2035. The smarter move is to understand what the tools can do today.

The venture capital firm 8VC concentrates on the practical application of AI. Its founder Joe Lonsdale describes the extreme optimism in Silicon Valley circles as having a religious quality. The assumption that more data and compute will bring about superintelligence ignores potential plateaus and technical bottlenecks. He prefers to view AI as a powerful tool for restructuring the back office of the global economy.

8VC lays out six stages of the AI wave. The structure is familiar to anyone who has studied previous technological revolutions.

First come chips and semiconductors. These are the GPUs and specialised hardware that train and run models. They are scarce, capital intensive and geopolitically sensitive. When electricity was new, this was the equivalent of generators and turbines.

Second comes compute and data centres. Vast physical infrastructure that houses the hardware and manages staggering power loads. In the electricity era this was the grid. Heavy investment, a slow build out and enormous fixed cost.

Third are frontier models. The large language models built by OpenAI, Anthropic, Meta and others. These are the equivalent of the early electrical appliances and motors that showed what was possible once power was available.

Fourth is the software layer that helps developers build, test, deploy and monitor AI systems. It makes the raw models usable in production environments. Transformers, converters and switches are the equivalents for electricity.

Fifth are applications that solve specific problems for businesses and consumers. Lonsdale argues that this layer will capture the most long term value. The value of electricity is most obvious in what it enables.

The final layer is services. These are the consultants and operators who map company data to workflows and design the art of the possible. While OpenAI or Anthropic provide the brain, enterprises cannot use it safely without a sophisticated middle layer that grounds the AI in the company's data and legal constraints.

The pattern mirrors railways, electricity and the internet. Infrastructure first, then tools and then applications. The final stage is widespread organisational change. For a small business owner today, all you need to understand is what applications and services can do for your workflows.

8VC argues that AI is good at five things.

Document processing sits at the top of the list. Modern models can extract structured data from PDFs, emails, invoices and contracts with high accuracy. A construction firm can pull line items from supplier quotes into a pricing sheet. An accountant can ingest client bank statements and categorise transactions before review.

To deploy this, identify one repetitive document heavy process. Test an off the shelf tool that converts those documents into a spreadsheet, or pushes them into your CRM. Measure time saved per week.

Browser and desktop automation is another capability. Think of logging into a supplier site, downloading reports, copying figures into another system and sending a confirmation email. Instead of hiring another administrator, a small business can pilot an automation tool that performs the workflow overnight. Start with a rule based, well documented process. Keep a human in the loop until error rates are acceptable.

Conversational and voice agents have improved a lot. They handle routine enquiries, support tickets and lead qualification. A local services company can deploy a website chat agent trained on its pricing, service areas and FAQs. A clinic can use a voice agent to handle appointment scheduling outside office hours. The key is to constrain the scope. Limit the agent to defined questions and provide escalation to a human.

Advanced research is a fourth use case. AI can gather information through multiple model calls and web searches, before synthesising findings. A marketing agency will generate a structured competitor landscape in a couple of hours. A procurement manager can compile well-sourced supplier comparisons. The adoption path is to formalise research briefs and run them through a research agent. Then validate the output against primary sources.

The final strength is synthesis and summarisation. Large volumes of text can be condensed into concise insights. Board packs, meeting transcripts, customer feedback logs and technical documentation can all be reduced to actionable summaries. This reduces cognitive load and improves follow through.

Each of these capabilities is incremental and a million miles away from artificial general intelligence.

The mistake many commentators make is to chase spectacle. They dream of autonomous agents running the whole business while ignoring the stack beneath them. Electricity only transformed factories when managers redesigned workflows around smaller motors and continuous production. The same redesign will be required with AI.

The opportunity today lies in disciplined application. Map your processes, identify friction and match it to one of the five capabilities where AI is strongest. Then pilot, measure and refine.

There is clear acceleration away from smart enterprises built around data retrieval. The AI wave focuses on automated decision-making and non-linear problem-solving. This is definite progress, but it still requires careful planning and implementation.

IDC reports that inefficient document processing costs businesses a collective $1 trillion a year, with the average knowledge worker spending 2.5 hours a day gathering information:

[1] 77% of employees in a survey by the Upwork Research Institute said AI increased their workload.