Your Next Customer May Not Be Human: The Agents Are Coming

The quiet revolution in automation is about to get loud

Automation Pipe Dream

Reasonably manageable task automation is not new - if you ever created a Zapier’s zap, you know how easy it is to string things together and end up with a cohesive flow of logic.

Zapier reduced the friction and cost of creating and maintaining custom script-based automation. Communicating with other software products through standardized interfaces points significantly reduced the brittleness of the process.

Yet the integration points must already exist or be custom-developed. And sometimes they still break.

What if automation flow was smart and capable of creating it’s own rules, self-adjusting and healing?

That’s the basic idea behind the Operator, introduced by OpenAI in January 2025.

What is Operator?

You may have already met Operator’s earlier and simpler cousins - Siri, Alexa, and Google.

You can ask Siri to search the web, add a new reminder and open computer applications. Alexa can order toothpaste and play your favorite song. The experience feels fractured (try ordering toothpaste with Siri), clunky and limited.

The Operator is the next step.

While Siri uses voice recognition to identify and execute the command (and responds with “I cannot do that” if command is not programmed), Operator uses reasoning, vision and ability to interact with digital interfaces like buttons and dropdown boxes to figure out how to complete your task.

Siri can only do what it’s programmed for and it’s limited by the device it’s on. Operator can figure out what it needs to do - across entire Internet.

Redefining the U in UX

UX (user experience) defines how people access your website or other digital products to purchase goods, receive information, or obtain services.

The interface is for people. But U (user) may not always mean “human user” - an intermediary autonomous agent could use reasoning to navigate and complete tasks on your site.

The future is here - almost.

A research published in spring of 2024 pitted humans against LLM in task completion. Humans won (79% success vs 14%). Less than a year later the published Operator benchmark showed significant improvements in task completion (between 39% and 79% depending on the task type).

And yet, watching the demos of Operator you can’t help noticing how clumsy and unsure it looks at times. It fails a lot, gets stuck, and asks for help.

Current criticism focuses on a wasted effort - who needs agents for such low-value tasks? But of course it’s just the beginning.

Consider just a few possibilities:

Use Case - Improving accessibility benchmarks

Benefit: simplifying resource usage for people with disabilities.

Accessibility is never done - it must be maintained and tested for. Due to complexity, cost, talent and business priorities the companies usually play catchup and as a result, some sites are difficult or even impossible to be used by people with disabilities.

Automated agents can provide assistance in navigating such sites.

Use Case - Bypassing complex legacy interface

Benefit: saving time by completing tasks in confusing web interfaces and reducing errors.

The application flows that have grown organically over the years can be expensive to rebuild. If you have ever tried to figure out an especially complex site where nothing can be found easily, you would appreciate the assistance.

Use Case - Reducing the cost and complexity of software development

Benefit: outsourcing simple tasks to agents, freeing up expensive developer resources for more impactful things, enabling business.

Imagine - instead of purchasing off the shelf software you can create your own just by asking an AI agent (you can - check out Bolt.new)

A family of agents

While Operator is the first of its kind, Bolt.new is another example of an AI Agent - an autonomous AI system that can understand, reason and complete tasks on behalf of the user.

You can use Bolt to create a web application and publish it - thanks to the website host (Netlify) which had developed a platform for AI agents to communicate with.

As more systems come online, imagine what the future would look like.

What is your plan?

You have enough time to prepare and be creative. Imagine the reasons.

Customer Interaction Automation

Automated subscriptions is a clumsy way of making it easy for customers to repurchase your product ( like Alexa and Amazon). But what if they could just tell the agent “hey I am getting low on fancy toothpaste, reorder it” - or better yet, an agent could reorder it for them?

Market Discoverability

SEO is here to stay but it may look different in the future. Imagine the world where Google search is replaced by agents looking for products and services on behalf of their customers. If your content is not properly categorized, it would be hard or impossible to find. Also consider new ways of cross-promotion and product suggestion.

Revenue opportunities

OpenAI is partnered with DoorDash, TripAdvisor and other selected few companies to accelerate the functionality. Bolt.new partnered with Netlify to host applications created by AI. Partnerships of the future may depend on your readiness to integrate.

Marketing and Catering to Users

Almost every business has a website now. Alexa reordering detergent would have considered magical twenty years ago. Future proofing your business means catering to the users of the future, taking advantage of innovative ways of reaching new markets.

What does it take to get there?

Regardless of your stance on where the future will take us, take an inventory of the effort required to agentify your company.

Website: clean house

Remember when mobile devices were introduced, and suddenly everybody was scrambling and adjusting their websites to be usable on tiny screens?

To make it easy for agents to navigate, review your navigation and site hierarchy and make use of aria labels.

Optimize data and flows

If you are selling products, standardize your required data tags. If you selling services, make sure it’s simple to request them - consistently.

Document your processes

Machines need scannable documentation to understand how to work with you. If you have automated interfaces, provide detailed documentation and examples. If you have a complex process, list all the steps in the FAQ.

Operations matter

If your business process is difficult or requires a manual step, now may be the time to review it for possible optimization. Can an entire request be completed online? Don’t forget about escalation! Having multiple ways of accomplishing something is confusing - for people and agents alike.

Design a cage

Safety first. Define a guardrail that agents cannot cross without handing off to the human - or where no agent may enter. Determine what is the reasonable amount of traffic and requests an agent can generate and if rate limiting would make sense.

Prepare your talent

The world where IT was quietly crafting solutions in the dark basement that nobody understood is over. It’s critical for business and tech sides to work together - training business people to speak tech and vice versa would be a secret competitive advantage of the future.

The future is coming

The digital transformation of the past is still applicable. If you pushed that project down the road, consider revisiting it now.