This is part two of Brand as Software.
When I joined Ramp ten months ago, my two senior brand designers were making decks and whitepapers. It made sense. They were the ones with judgment and taste and craft, and the company needed those assets. But I wanted them to build the brand system, preparing us for what turned out to be an incredible year of growth and new products at Ramp. So I took them off the decks and whitepapers right away. Contractors took over the production, and my designers got their time back.
What I want from AI, and what I want it to give creative teams, is time: time for deep work, time to let ideas simmer and marinate, and time to go deep on the things AI cannot do well. Where we give our attention, and how we spend our time, will determine the quality of our ideas and execution.
Time is scarce. We have ancient brains, and weāre using an alien technology that pushes our attention further every week. Most of us get three or four hours of solid concentration a day before the brain starts to fizzle. Parkinsonās Law says work expands to fill the time you give it, and sometimes what you think takes a month takes a week. But I bet a lot of creative teams are one more week, or one more month, away from taking their best idea to the next level. Whatās stopping them is the constant urgent requests from every team.
When your creative team is making the same assets over and over, itās wasted talent. When you pair creatives with builders who codify the teamās quality and execution, especially for repetitive work like documents and decks, you can scale creativity and protect the rare talent that makes a company stand out.
Hereās how we set up brand as infrastructure at Ramp, what weāre learning, and what the time we got back made possible.
Cover the production, protect the time
Like I joked in the last essay, the more I use AI, the more I see the past of my career as laughable monkey behavior: hunched over a keyboard, pressing CMD+C and CMD+V from one rectangle to the next. Itās silly to call this āwork.ā Give it to the machines and get back your time.
Bringing in contractors was the first step, and it was a staffing decision before it was an AI project. If I had asked my designers to evolve the brand system on top of an unchanged production queue, I would have given them a second job, with no time to go deep on anything. The production still had to get done.
One of the designers we brought in, Kyla, showed me the possibility of how designers can use AI. By her second week, she had used the AI tools at Ramp to build a tool that automated about 90% of our whitepaper work. Then she came to me and asked: that team has the tool now, what do you want me to work on instead? Thatās the future. The second step was making sure that production work didnāt disappear once it shipped. Normally a deck or one-pager has the shelf life of a banana. We now have the technology where the good decisions and good output get codified into the agent as skills, guidelines. We create multiple 10/10 examples in Figma, so that the agents can learn what approved, good work means to the brand team. It has to pattern-match excellence.
It helps to be precise about which work this covers. There are decks that agents can create so anyone at the company can self-serve. Others need to be hand-crafted, like the one for your company conference. The hand-crafted work still belongs to designers, and it should get their full attention. The best part is what happens when that deck is done: its quality gets added back into the brand system, where the agents can see it. Now you have a new template, new animations and new storytelling that are part of your brand infrastructure, and the standard of good work elevates everyone.
Thatās brand as infrastructure. Designers make creative decisions. Builders turn the repeatable ones into tools the rest of the company can use. The designersā time goes to the decisions nobody has made yet. They have time to invent the future, imagine, and dream. Rather than having a short turnaround time to execute good work, they have more time to push the idea further.
I believe protecting this resource, nurturing it, will help companies stand out and win the marketplace.
Judgment is human. Repetition is software.
The Michelin-star vending machine
At Ramp, we have an internal Slack agent called Ramp Creative that makes on-brand decks, whitepapers, one-pagers, print material of all kinds, and product graphics. It can also edit images, like changing a $ to ⬠or swapping a customerās logo. Weāre working on getting it to make product demos, reels and videos too.
The person asking for a deck wants something they can use, edit and refine. Humans still desire control. They shouldnāt have to learn the finer points of typography, hunt through three libraries, or become an expert at prompting before they can do their job.
I quickly learned that itās not about making a bunch of tools for everyone. Tools are personal and should be an extension of the person with expertise.
What everyone wants is a Michelin-star vending machine: press a button and get something excellent. That means the brand team does the hard, strategic work before the requester ever places an order.
Most creative teams have run like a sandwich counter. An order comes in, the team puts the sandwich together, and it goes out. The best creative teams of the next decade will run test kitchens, with the time to let ideas marinate, to experiment, to find out what works and what fails. Their best experiments go into the vending machine.
A fair question is whether this system leads to settling for good enough. Quite the opposite. I avoid the word ātemplateā because people hear lower quality: itās just a template, so whatever. A Michelin-star vending machine only works if what goes into it is the teamās best work.
All the quality work we shipped. Now it goes back into the system.
Hereās how a request moves through it. Someone asks Ramp Creative for an asset in Slack. The agent picks an approved skill, which is a set of instructions and rules for that format, including what can and cannot change. It fills the layout with the requesterās words and approved images, a renderer turns that into the finished file, and automated tests check for broken dimensions, missing fonts and overflowing copy before the asset comes back in Slack. Behind all of that, the brand team designs and approves the 10/10 examples in Figma, the skills live in Dojo, our internal library for sharing them, and the agent runs on Inspect.
When someone asks for something the agent doesnāt have a skill or template for, it vibecodes an attempt. We set up an automation that identifies an asset missing its core ingredients, adds it to a database, and pings a designer. The designer has 24 to 48 hours to turn the vibecoded slop into 10/10 examples of that asset, so the agent can pattern-match better. Then a skill formats the Figma file so itās ready to encode into Ramp Creative.
I realized the bottleneck as I was writing this. A designer still has to make each new 10/10 example by hand, even if a senior designer approves it in under a minute. Next, the agent makes 30 versions in Figma, and the designer only comments: this works, this doesnāt. Approve the work, get it in the system.
For the formats we support, 80% isnāt good enough. An agent that gets people most of the way repeats the same bullshit weāre trying to escape: people generate something, knowing a designer is there to save the day. If this agent canāt give people 100% of what they need, Iād rather hire 10 production designers cranking out finished assets every day.
Reliability is the price of entry. Iāve learned that if the agent doesnāt deliver within a couple of attempts, people work around it, prompt their own AI, and the brand-as-software philosophy fails.
The UK launch
Sabrina from marketing pushed Ramp Creative to produce and localize more than 80% of the assets for ourUK launch. The brand team had already made the design and creative decisions. Sabrina could self-serve the work she needed, from product graphics to decks to documents.
Before, the brand team would have had to compromise. Do we work on the UK launch because itās the most important thing right now, or do we work on the conference or the brand or the 100 other urgent requests? Thatās the tradeoff this system helps us avoid. This is the challenge of nearly every brand team from the past two decades.
OnRamp Conference
OnRamp is @tryramp yearly conference for finance leaders and its biggest brand moment. In 2026, it welcomed more than 1,200 people to San Francisco and 2,500 more online, and itās where Ramp showed its product roadmap, including the first public look at Glass.
If you've ever had to do events, there is never enough people or time. Last minute requests, no matter how much you prepare, are normal. So, we used the agent to make a lot of edits, fixes, and assets completely self-serve.
People ask me how, in my ten months here, we pulled off a full brand refresh in four weeks and branded Ramp Labs, Stack, Glass, Inspect and Router while Ramp was also shipping Bill Pay, Accounts Receivable, international expansion and other projects. Each of those needed landing pages, assets, demos, copy, videos, and product graphics.
One answer is talent, which Ramp is overflowing with. Second, I am insanely grateful for our agency partner, Common Reality, and my team for focusing on execution. Overhauling the brand system in four weeks enabled everyone to do better work. We had fewer reviews with other teams because of it. Itās not that my team didnāt want to work on core launches like Bill Pay or UK Launch; it was really about opportunity cost, and new flagship products like Glass and Inspect allowed us to keep pushing the edges of our system.
One last example. Sophie, from the SEO team that works on Router, uses Ramp Creative thoroughly. She self-serves everything she needs from new launch assets to landing pages to product graphics and social posts. Internally, we look at this agent as a product, with customers that we need to take care of. As new assets are needed, the system gets smarter because it pattern-matches to other work. The goal is to keep feeding back good layouts, compositions, and work to the system.
Deep work needs an environment, resources, and protection. Brand as infrastructure changes the DNA of a company. And for too long, creative teams just werenāt allowed to do this or had to burn out and white-knuckle it all. That isnāt sustainable, especially for the era weāre entering where anything can be built by anyone.
Artists, builders, and brand designers are the rare talent that will make your company stand out.
What would you do with more time?
One of my favorite projects at Ramp isInspect, a cloud coding agent that the engineering team built internally. It handles over 80% of our PRs, and itās one of the coolest things Iāve seen and used. We opened up beta access on October 27.
When I started on the launch, I found out that Sabrina had already spoken to prospects and design partners. I interviewed Nico, the engineering lead, and his team. I also interviewed Marcelo, a product designer who loves using Inspect.
In the old world, this is where meetings about meetings start: endless documents, long Slack messages, stakeholder reviews, 50 comments to address in Notion or Docs, and too many chefs added into the kitchen. Because there was conviction and clarity on what Inspect is and who itās for, we skipped most of that theater.
Iāve been honing a bundle of skills that take me from research to a brief, to brand strategy, to a campaign platform, and I used them on everything Sabrina, Nico, Marcelo and the team had told me. I put the brief together in minutes. I got Sabrina and Nico aligned in a few days. I briefed the designers by the end of the week. Three weeks to come up with multiple, strong directions is intenseĀ ā plus landing page, copywriting, and assets ā but thatās how we operate, and we were able to keep the quality high.
Why? Because itās not that Iām relying on AI to give me an answer. Iām using this tool to help access parts of my past, my career, and my brain. I can tell quickly whether something feels right or wrong. People call that taste, but you donāt award yourself taste, any more than you award yourself a Michelin star. Itās earned after years and years of hard work, and almost arrives late. What you start with is a point of view.
The brief drew on Sabrinaās customer conversations and the people who build and use Inspect. I sent Nico a dozen questions in a Google Doc, and his engineers answered them. I also scraped every past conversation and Notion document about Inspect. The agent came back with three brand strategy options.
The AI companies scraped the internet. They didnāt scrape my Dropbox: decades of client work, decks I rewrote and rebuilt, thousands of essay drafts with comments from editors that refined the work, and 60,000 images Iāve saved over my career, now indexed and tagged. From those, I pulled six art direction worlds to explore with the team. Speed comes less from the technology than from the person with a point of view able to bend the tool to their needs.
It's like looking at a design portfolio: it only takes me a few seconds.
What I care about is what we did with the time. Sabrina, Nico and I got to banter and question the messaging, language, framing, words and outcomes. I got to really understand why Marcelo loves using Inspect as a product designer and learned how other people in the company use it directly in Slack.
The brand team had more time to explore directions, to question how Inspect connects to the Ramp brand system overall, and to push the work. People are probably realizing that the bottleneck to excellence was always the issue: too many chefs in the kitchen, but more importantly, the wrong chefs. On the Inspect launch, a few experts trusted each other to do their jobs well. Collaboration, trust, and respect for craft will never go out of style.
So many good ideas die because teams thrash at the end instead of the beginning. If everyone thrashes in the first 20%, the bad ideas come out early, and thereās time left to execute the good one.
What leaders have to change
The deeper problem is how decisions get made.
Giving everyone more tools can intensify an old problem: the ability to generate something becomes another way to overrule the person responsible for making it good. Thatās why I think brand teams should be set up like security or infrastructure teams.
The comparison has limits. Bad design doesnāt sink a business overnight the way a breach can. The damage over time is slower: one year you ship the new rebrand, and by next year, you look like three different companies. With AI, people will generate random assets, thinking it looks good, or worse, āfine,ā and ship it. Thereās no way for a brand team to monitor everything that goes out the door.
Design and creative output in tech is treated like movies or fashion. Everyone can have an opinion about it. Ask any professional writer or photographer and they know what Iām talking about. Because people can read, they think they can write. Good writing makes it look easy, and now that AI makes production look easy too, itās tempting to believe output can stand in for skill, taste, and experience.
What I want to borrow from a security team is the way companies fund and trust it. Security gets dedicated people, clear ownership, a budget, the authority to say no, and accountability when something goes wrong. If a brand team is made up of artists and builders, and treated that way, what itās accountable for is simple: make the brand easy to love and impossible to misuse.
In a more truthful tone: no one tells the security team to go fuck themselves.
At Ramp, that now means Ramp Creative has a dedicated backend engineer and a designer, and Iām in the codebase myself. To be honest, a lot of our best work from this year isnāt in the system yet. Thatās what the people are for.
But historically brand and creative teams are at the end of the line. The last to know and the first to get blamed. When there is no clarity or strategic thinking from the people who start a project or need something from the brand team, the creatives have to fix it. They have to act as detectives, journalists, and therapists. And they have no time to go deep. In many ways, brand and creative teams have been treated and spoken to like LLMs for decades. All of that is about to fundamentally change.
AI can make the sandwich counter faster, or it can give the team a test kitchen, and leaders decide which. Fund designers and the builders beside them with committed time. Cover production while the system is built, and settle upfront who makes the final creative call. Then protect the time you get back so the team can make better work and feed its best decisions back into the system. Everyoneās work starts from a higher standard.
The north star is simple: Incredible craft at incredible scale.
None of this replaces collaboration. I love working with PMs, PMMs and engineers who respect what I do, cheer on good ideas, and share signals that help me adapt the idea. If they have time to talk to customers and gather signals, thatās exactly what my team needs to begin. I can scrape the docs and Gong calls, use a set of skills to form a brief, and then we can spend our time together debating the brief and thrashing early.
I bet youāre moving faster, but not going deeper, and still looking to Twitter and other companies to borrow their success. A company can keep growing while it becomes boring and unimaginative. Success hides failures. The ZIRP era is gone, you canāt throw money at the problem anymore, and I think employees and consumers will be quicker to leave brands they no longer believe in. You need a story, a point of view, and the execution to make it real. This is where competition is going to get spicy, and I want more of it.
Build the infrastructure now
I donāt have a crystal ball, but Iād rather we build for next yearās landscape than todayās. This is also a great moment in time to connect brand to real business outcomes. For my entire career, āmeasuring brandā was a moving target. Sure, share of ābranded searchā or brand awareness is measurable, but it's one of many metrics, and doesnāt always paint the full picture. Meta and Google are the tails that wag your strategy and spend.
Now, brand can prove that this agent creates a deck that costs, maybe, a dollar, versus $100 an hour for a person. This isnāt about replacing humans; this is about putting your best creative talent on hard problems instead.
Itās all about opportunity cost: if you werenāt doing this, what would you be doing instead? Brand design is arguably the rarest talent in tech; there are plenty of engineers and marketers. Why, then, are they spending hours fixing other people's mistakes? This team is about inventing futures that no one is thinking about.
I recently connected with our SVP of Sales, Max Freeman, and audited how his best sellers work. From how they make decks, what they prompt, how they sell ā this is a design challenge. Empathy, user experience, and imagining what better looks like. Now we can take all of that work, get the best brand talent on it, and level up everyone else in the org. The agent levels up. The system introduces high quality at scale for everyone. Great design is not an asset, it's what it enables people to do with their time.
Start with decks, whitepapers, and the small tasks that add up, like resizing logos or swapping customer logos. These are important for a business, but they should not be eating up your creative teamās time. Itās like having Olympic athletes refilling water bottles.
Pair the brand/creative team with engineers. In essence, you are building a software product. It cannot suffice as a tool. It has to become a system of tools, agents, skills, libraries, guidelines, recipes, evaluations, and tests.
Imagine youāre a startup and you have a fresh new brand. This infrastructure can help every new hire represent the brand well in their work. Imagine you rebranded a year or two later. Well, you feed that work back into the system, and the entire score of music updates. Rather than taking months applying the new brand to every channel and surface, you treat the brand like a software update.
This is all very very fun for me. Itās like watching a history book evolve in real time. I was always confused by how companies with so much potential and money operated. I understood all the drama that tech spews on Twitter about every function or team. AI is the wrecking ball for this house of cards, these illusions weāve been defending for years.
Companies need to fundamentally reimagine how they operate. Careers are going to be wildly different. Job titles will get ever more confusing. The only thing that matters right now is adaptability and execution, kindness and collaboration.
Many companies make noise, but who will make music?
I believe that begins and ends with your brand and creative teams.
More power to them.
A heartfelt thank you toCaroline Donofrio for editing this essay, and to Mint Zhai, Eleanor Ngai, and Stephen Ong for the beautiful demos that brought it to life.
Massive thank you to Emerline Ji, Drew Minns, Burak Aslan, and Mary Jiang for making Ramp Creative possible; I could not have done this without you.
Enormous gratitude to Mint Zhai, Kyla Arsadjaja, Claire Chen, and Coco Yuan for your 10/10 design work, and to Mark Weaver and Aodan Reddy for keeping the bar high.
Thank you to Diego Zaks for believing in this endeavor, allowing me to fail and fail until we got it right.
Special thanks to dear friends who read early drafts and provided useful feedback: Behzod Sirjani, Mike Smith, Sean Blanda, Lucas Crespo, Flora Guo, Elizabeth Lin, Katie Siu, Mark Johnson, and Michael Casebolt.
And please check out my latest podcast episode with Ridd from Dive Clubwhere I share the demos, tools and thinking behind this essay in more depth.






