I Built AI Agents From Scratch and It Cost Me More Than I Expected

I told myself at the start of this year that I was going to build something using AI. Not just use AI tools, but actually build something with it. I kept seeing people online make it sound simple enough that almost anyone could do it. So I figured, why not me?

What started as curiosity became a proper rabbit hole.

How It Started

It began with a YouTube video. I uploaded it as a reference point, a demonstration of what AI agents were capable of doing through Telegram. That video became the seed for everything that came after. I watched it, took mental notes, and thought to myself, "I can do this."

Well, I could! But not without a few surprises along the way.

Building Natalynn and Personal Assistant

What I did not expect was how quickly things could come together when you actually commit to it. In less than three hours, I had built working agents from scratch. Three of them, actually. Natalynn (client facing agent), Personal Assistant (response only to me - does my scheduling, search, read, draft emails and documents), and an Admin Agent (to handle more complex tasks). Not prototypes. Not demos. Actual, functional agents doing actual things.

That part genuinely surprised me. The tools available today make it possible to go from zero to something working in a sitting. I used Claude Code running through Windows PowerShell, set up an Ngrok server, and learned how to wire everything together working on .env files and API keys. None of that was in my vocabulary few weeks ago.

It was a real learn-by-doing experience. You figure out what each piece is for, why it matters, and how they connect, because nothing works until they all do.

Natalynn negotiating meeting times and booking it after checking with me

Personal Assistant booking my calendar and giving dinner suggestions

What Natalynn Can Do (for now)

Natalynn operates through Telegram as its primary interface. When an external party reaches out, the agent reads the intent of the message and routes it down one of three paths. For appointment requests, it checks my calendar for availability, proposes times directly to the external party, handles any back-and-forth negotiation autonomously, and only loops me in to approve or decline a finalized slot before confirming. 

For service enquiries, it presents the full list of services with pricing, collects the person's contact details as a lead, and issues a quotation if they express interest in moving forward. Beyond business, if the person is in the mood for something lighter, the agent can throw in a joke. The whole point is that I only touch the workflow at the one moment that actually requires my decision, and everything else is handled automatically end to end.

Natalynn & Personal Agent workflow (Click to expand view)

The Part Nobody Talks About Enough

Here is where I want to be straight with you, because I do not see this discussed honestly enough.

It is expensive.

Not in a casual, "oh you might spend a bit more than expected" kind of way. I mean genuinely expensive, in ways that sneak up on you.

Every time you make a change to a workflow, add new functionality, or fix a bug, you are using tokens. Every query your agent processes uses tokens. Every step in an automated workflow uses tokens. Even getting a simple reply back costs tokens as it hits the API. It all adds up, and it adds up fast.

The more you automate, the more steps a workflow takes, the more complex the task, the more it costs. Multiply the number of queries and tasks your agent handles per day across a full month, and you start doing some uncomfortable mental arithmetic.

Which led me to a question I did not expect to be asking myself after building all of this.

Is It Actually Worth It?

At a certain point, the cost of running agents and automation at scale starts to look a lot like the cost of hiring an actual person. And here is the difference: with a person, you can control the cost. A monthly salary is predictable. Token usage is not, especially when automation complexity grows over time.

What about you? Have you built an AI Agent or automate a workflow before? Would love to hear your experience.

The Karma Swoosh

If you have followed my last few posts, you know I have been tracking Nike's slow slide from a marketing perspective. The Nike vs. New Balance story was about momentum. This one is about something more uncomfortable: originality, or the lack of it.

Nike got sued for these Air Max.
Photo credit: Nike

On 1 July 2026, 7-Eleven filed a lawsuit against Nike in the US District Court for the Northern District of Texas. The claim is that Nike's upcoming Air Max 95 colourway copies 7-Eleven's tri-colour mark, the orange, green and red stripe combination the convenience store chain has used for nearly four decades. To make things worse for Nike, the shoe was set to launch on 11 July, which is 7/11, the same date as the brand's own "Free Slurpee Day". That is not a coincidence anyone can explain away easily. 

The 7-Eleven Lawsuit, In Short

A few details from the filing stood out to me as both a marketer and a sneaker collector.

•      Three colours, orange, green and red, are all it takes for most people to think of 7-Eleven the moment they see them together. That is the real story here. It is proof of just how strong 7-Eleven's brand equity is, when a colour combination alone, with no logo or wordmark attached, does the job of instant recognition. Not every brand can make that claim, and it is exactly why the lawsuit has legal teeth.

•      7-Eleven says product listings and media coverage had already started calling the shoe the "7-Eleven" sneaker, before it even released. That is the kind of unplanned brand association that should set off alarms in any legal or marketing review.

•      The complaint claims at least one consumer bought a pair believing it was tied to 7-Eleven. In trademark law, that consumer confusion is the whole ballgame.

•      7-Eleven is not asking for a slap on the wrist. It wants an injunction, a recall and destruction of the shoes, disgorged profits, and damages that include treble and exemplary damages plus legal fees. Nike has already pulled the Air Max 95 from its SNKRS app in response. The case is filed as 7-Eleven Inc v. Nike Inc, in the Northern District of Texas, case number 3:26-cv-02201-X.

•      Colour trademarks are genuinely hard to defend in court, but they are not impossible. The Supreme Court's 1995 ruling in Qualitex Co. v. Jacobson Products Co. established that a company can trademark a colour if it can prove consumers recognise that colour as belonging to its brand. 7-Eleven's entire case rests on proving exactly that, and given how instantly recognisable those three stripes are, they may not have to work too hard to prove it.

The Irony Nobody's Talking About

Here is what makes this story worth writing about. In March 2026, Nike won an 11 million dollar verdict against Divide The Youth and its founder, social media influencer Nicholas Tuinenburg. Nike had sued them back in December 2023 for selling "Division Dunks", sneakers that a jury found copied the design of the Nike Dunk Low, right down to the silhouette. Nike argued this created consumer confusion, the exact same argument 7-Eleven is now making against Nike. The jury sided with Nike and awarded 8 million dollars in statutory damages and 3 million in punitive damages.

So the brand that spent years in court arguing that copying a shoe's design and trading on its recognisability is worth 11 million dollars in damages is now the one being accused of doing precisely that to someone else's iconic branding. If 7-Eleven's version of events holds up, Nike would be found doing to a convenience store what it once dragged an independent streetwear label through years of litigation for doing to them. That is karma, and it is hard to write a cleaner case study of it.

This Whole Thing Could Have Been a Collab

Credit where it is due, even if entirely by accident: had Nike simply picked up the phone and pitched a follow-up collaboration instead of quietly building a similar colourway on its own, the timing could not have worked out better. A launch on 11 July, aligned with 7-Eleven Day and Free Slurpee Day, is the kind of date a marketing team would beg for in a real partnership. Instead of a feel-good nostalgia moment building on a collaboration that already had proof of concept, Nike turned a gift of a launch date into a federal lawsuit.

Actually, back in 2020 both 7-Eleven and Nike had collaborated on a Nike SB Dunk Low based on the 7-11 theme. Unfortunately, it got cancelled due to mixed review of the colorway and design prior to release.

Nike x 7-Eleven SB Dunk Low collab back in 2020 that got cancelled
Photo credit: Sneakernews.com

Why This Matters More Than a Lawsuit

I keep coming back to a theme from my earlier post on the sneaker rebalance. Nike's innovation gap runs deeper than product, reaching into culture and originality too. A brand with the design resources of Nike should not need to lean this close to another company's protected branding to generate hype around a launch date. When New Balance wanted heat, they handed creative control to Teddy Santis and Joe Freshgoods and let them build something new. Nike, in this instance, allegedly leaned on borrowed recognition from a convenience store chain and a calendar coincidence.

Whether or not Nike wins this case in court, the story itself tells you something about where the brand's creative confidence sits right now. A company that used to set trends is now being accused of riding someone else's.

Key Takeaways

•      A colour alone can be a brand asset worth defending. 7-Eleven's tri-colour stripes work without a logo attached, and that level of recognition is rare. If your brand owns a colour, pattern or shape this distinctly, treat it with the same protection you would give a wordmark or logo.

•      Brand equity cuts both ways. If you have spent years litigating to protect your own designs, expect the same standard to be applied to you. Consistency in how a brand treats intellectual property protects its credibility, not just its legal position.

•      Unplanned consumer confusion is a red flag. When people start calling your product by another brand's name before launch, treat that as evidence a legal team could use against you, rather than as free publicity.

•      Originality is a renewable resource, but only if you invest in it. Nike's innovation gap was already visible in slowing growth and franchise fatigue on the Dunk, Jordan 1 and Air Force 1. This lawsuit adds a new dimension, with the brand now accused of borrowing someone else's equity rather than building its own.

•      Reputational risk compounds. A single lawsuit is a legal matter. A pattern of lawsuits, especially ones that highlight contradictions in a brand's own past legal arguments, becomes a brand story that outlives the case itself.