The Impacts of Agentic AI with Amir MXT

Few terms are as buzzy right now as “agentic artificial intelligence (AI)”. For Amir MXT, co-founder of Humblytics, an AI native analytics and conversion optimisation platform, agents are not an abstract possibility, but a concrete part of his day-to-day workflow. 

While they are still in the early days of implementation, AI agents have the potential to revolutionise not only the speed and productivity of companies, but how work is done on a fundamental level. Recent analysis of how people use Claude Cowork has shown that agents are overwhelmingly deployed in software engineering contexts, but could soon play a much larger role in other fields such as marketing, copywriting, sales and accounting. 

Amir has built a career advising clients on how to deploy AI agents using the tools currently available. When consulting, Amir encourages founders to approach the experience with a key question: “How do I get to $1m-2m in revenue, or more, with as few people as possible and leveraging a lot of AI tools?”

Deploying agents can shrink the timeframe and resources required for a company to scale effectively, and Amir has seen the results himself. “Previously we had a business with about 17 people, doing about $3m-5m in revenue every single year. Now we’re on track to hit those targets with a much smaller team.”

In the latest episode of Opto Sessions, Amir unpacks how to treat AI agents as a co-founder, why LLM readability is becoming as important as SEO, and why, for some, AI implementation might mean working more – at least in the short term. 

Meet Your New Cofounder

When LLMs first came into use, AI was primarily used to aid individual workers in the completion of general use tasks. The rise of OpenClaw in late 2025 changed that, with the popular open-source GitHub project allowing users to create AI agents to carry out tasks autonomously. Now, Amir explains, a wider range of tools allow “augmentation and agency, where AI is fully embedded into the core workflow with the human still in the loop, and it can serve multiple functions.”

Since then, Anthropic has emerged as the current leader of agentic AI, Amir says. “They are solving for problems that we didn’t know existed.”

Creating and implementing AI agents can seem daunting, especially for people without a tech background. However, Amir notes, the advancement in tools means it has never been easier to lower that learning curve through trial and error. 

When working with clients, Amir says, “the friction point of a terminal was so scary to them, but now it’s not anymore… I would just say the best thing to do is download Claude Code, come up with a specific idea or problem that you’re trying to solve for, and then work backwards. Learn and understand how it’s approaching solving the problem and just build something… You learn by doing.”

That is changing the nature of computer-based work, he notes. “I used to say cursor was the interface for work. I take that back now. I think the terminal is the interface for work.”

Before clients begin that process of trial and error, however, fully understanding and mapping out the work they want to automate helps to ensure the effective deployment of agents. “It’s very workflow-specific,” Amir explains.

The deployment of AI agents means that the role of humans has changed as well. 

Just as more experienced staff might supervise a new hire, AI agents require human oversight to ensure tasks are accurately carried out, and that the output is aligned with company goals. Amir recommends asking key questions as you go: “Are you able to delegate tasks to it? Are you able to define what the output of that looks like? Is the model able to actually be aligned with what you’re looking for? Is there discernment? And is the information accurate?” 

Most importantly, “are you able to take full responsibility for the output that’s coming out of it?” 

Learning the Language of AI

As AI models are used for more and more applications, ensuring your output is optimised to be understood by LLMs is a key step — one which is just as important as SEO. Readability thus becomes a key concern. “The future is probably going to be agents interacting with services and websites just as much as humans,” Amir explains.

The process is already in full swing. “We’re seeing a lot of infrastructure for authentication, payments and email.”

He cites Stripe’s recently launched Machine Payments Protocol, which is an open-source, internet-native standard designed to enable autonomous “machine-to-machine” commerce, and Cloudflare’s [NET] crawl endpoint, which allows models to crawl entire websites with a single API, enabling more efficient and compliant training, as well as researching and content monitoring activities for agents. 

So, for developers, the question becomes, “how do you build interoperability between all these different systems to have a readable format between them to be able to exchange information? How do you create a new common language?”

In his own workflow, Amir already has an answer. “A common denominator between a lot of the models is that they interface with JSON, text files and markdown files.” He recommends Obsidian, a note-taking app that creates locally stored markdown files that can easily serve as context for AI agents. 

The first step to building agents is creating a library of context files, which the agent will reference to produce more consistent outputs. “I use AI to build out a set of context files that I would use within Claude Code to say, ‘reference this context folder, so you have a full understanding of our business and what we do.’”

Understanding which tools fit into each part of your workflow is key, Amir explains. He provides an example for an automated email outreach campaign. “So what I typically do is use Claude Code to build out a strategy. Then from there, I have it mapped out on Obsidian, and then go and execute it in Hunter.io.”

Productivity Gains

One key argument for adopting AI is its role as a productivity multiplier. It does increase productivity, Amir says, but that doesn’t mean that top users of AI are working less. “It’s interesting. I have a higher output than I did before with these AI tools. It was supposed to save me time, but, actually, I’m working more.”

Part of it is the new opportunities opened up by collaborating with AI, and part of it is the break-neck pace of the field, where the main competitive advantage is staying ahead of the pack. In the foundation model race, for example, “I think there is no moat,” he says. “The moat is speed, execution.”

Few people would question that AI is changing the nature of work. However, how it will change depends a lot on how people use AI tools, including agents, today. “What we’re building with our companies is a test bed for what we think is going to be the future.”

Even as someone pushing the future of AI’s role in work to its cutting edge, Amir sometimes has his doubts. “It’s existential crisis for me… am I going to be relevant today or tomorrow? And maybe I have a two or three-year window to just make as much money as possible and then go live on a farm.”

Still, the future will likely be somewhere between a tech utopia and a AI-slop-riddled dystopia – and where we end up depends a lot on what we do now. In that sense, Amir encourages anyone interested in AI adoption to play around, to see what works and what doesn’t. “I’m still experimenting. But I want to show what’s in the world of possibilities.”