My professional career started when I was 12, I started building websites for folks on the internet. We’d find each other through forums, they’d pay my invoices, never met in person. At 14, I accessed a router via my play-station-portable, accessing an unprotected AS400 system on a network at a medical billing service and pulled medical records. I showed the staff and they offered me a job in their IT to fix their network stack.
In college and after, I worked across ML, computer vision, and large-scale engineering—at Caterpillar, in research labs, and later in R&D at Capital One. I led teams, built ML platforms, shipped early generative AI products, open-sourced major projects, taught deep learning to thousands of engineers, and published along the way.
I’ve always viewed myself as in a trade, I was a programmer. I learned it, practiced it, mastered it, and helped pass it on. But that trade is now in its twilight. Some parts will remain, but programming as we’ve known it is ending—marking a moment of excitement, optimism, and nostalgia.
In a sense, AI has or will soon make it possible for anyone to become a content creator of any kind – writer, musician, programmer, videographer, designer, marketer, and yes… patent drafter.
AI Engineering at IP Copilot
One thing we’ve been working hard at during 2025 was trying to integrate automation as heavily as possible into our engineering workflows. In the first half of the year, that was mostly using AI on small sections of code and ask basic / small questions and doing security checks.
The real step change came in May 2025 with the release of Codex Cloud by Open AI. We lightly tested it for 6 weeks and in July we pushed it fully into our workflows. Below you can see how quickly it’s helped us accelerate our pull requests (PRs) on one of our repositories, each PR are in effect one features added or bugs fixed.
Since we’ve started using Codex, with the same team, we’ve increased our PRs by 2.5x and the code volume by 5x. You may be asking yourself, is the AI generating unnecessary code? Yes, it is. However, as noted by the team, it used to take 3-5 PRs to get a major feature out and now we’re doing it in one. If we have an additional 20, 30, 50% code who cares? Ultimately it’s results that count.
Every week I’m in a meeting with a prospect or customer, we see a minor issue or something we can improve. We log it all, take screenshots, and send it to Codex at the end of the call. I then log the issue, the PR, the call transcript / video and send it to the engineers to validate. We can get a release out same day.
When the Code Learned to Sing
With this acceleration in development our main bottle neck now is code reviews and even that I expect to diminish over time as the AI improves. As a team, our goal is to remove every bottle neck and we are getting increasingly good at piping requisite data to the AI at the same time the AI is getting increasing context and performance.
What does this imply? We made a song for you below —
Within the next 4-6 months I expect to look back and say “wow, I haven’t opened up my development environment even once”. It’s what I’d call a step change, where an entire form of work is now changed beyond recognition. In this case, programming will become largely obsolete in 2026 – 2027. We’ll still switch to manual mode periodically, but already 70% of our code is AI generated and increasing.
AI Design at IP Copilot
Around August–September 2025, our engineering velocity exploded as teams started burning through issues with Codex. Almost overnight, design became the bottleneck. Bryant, our Head of Design, is exceptional—but when engineering output doubles in a month, even the best designers need leverage.
Bryant started searching for AI-native design tools. We’d been using Figma for over a year—it worked well, but had no real AI-driven design workflow. Adobe offered generative assets, but not end-to-end product design. That’s when he found Lovable: an AI design platform that lets you chat your way to a working application.
Since September 2025, Bryant has increased his output by 10–20x. We can now put functional mockups in front of customers within 24 hours. These aren’t static designs—the AI generates real interactions and flows, producing a working application rather than disconnected frames. That clarity dramatically accelerates engineering, and the output from Lovable can be fed directly into Codex for rapid iteration.
What are the results?
Lovable has provided an AI army of designers and engineers at our Head of Design’s fingertips. According to Lovable he’s a top 1% user, it’s magical, powerful and getting better every day.
Let me render, one more time
Within 3 months, we’ve dropped Figma. Design tickets are closed within a day, not a week or two. Functional applications are provided to engineers to build off of. Design is no longer the bottle neck and ultimately Bryant’s tasteful touch is coming to life almost instantly.
Design has become simply describing what you want, iterating, waiting a few minutes and BAM. Functional app. Anyone can now copy or launch a competitive application within a day or two that is complex for a few hundred dollars.
What do design and engineering have to do with patent drafting? Everything. In 2025, AI moved from basic bug fixes to fully managed pull requests—tests, security scans, and reviews included. Design followed the same path, evolving from asset generation to building complete, functional applications.
Frontier models are already capable of drafting patents. The common objection isn’t that they’re “not good enough”—it’s that the workflow is wrong.
The real breakthrough in our use of AI for engineering and design came from workflow, not model quality. Tools like Codex let us queue work, run jobs in parallel, iterate quickly, and review continuously. That leverage—not raw text generation—is what unlocked 10–100x gains. We expect the same shift to define patent drafting.
Today’s patent tools still rely on accelerating typing or editing text. That approach mirrors early AI coding assistants, which delivered incremental gains at best. Real transformation won’t come from faster text—it will come from agentic, parallel, reviewable workflows.
Everybody’s got a claim
2026 will usher in a new paradigm for patent drafting—one closer to tools like Codex or Lovable than today’s text editors. A workflow-first, agentic user experience that actually makes sense. It might not be IP Copilot that builds it—we’re not currently developing a drafting tool—but someone will. The interface is inevitable.
What happens then? We even made a song about it (my personal favorite song on here):
Once the right user experience exists, the industry will change quickly. The exact timeline is debated—six months, 12 months, three years—the order of magnitude is months, not decades. We know the shift is coming.
Here’s the uncomfortable truth: this is a step change. Once an AI can iterate on a patent draft and review it, patent drafting becomes accessible to anyone for a few dollars. Just as I can generate a song in a few minutes today, anyone will soon be able to generate a patent.
Is that the end of patent attorneys? No. But it is the end of twenty-hour drafting engagements. Drafting, like music creation in this article, becomes something anyone can do.
What remains is the real value: strategy, portfolio design, licensing, competitive positioning, and identifying what’s worth protecting. Drafting itself will move in-house, driven by speed and cost. Law firms will need to adapt accordingly.
Why we built IP Copilot
IP Copilot’s vision has always been simple: unlock the innovation already inside your company. Capture it, protect it, and make it usable. In effect, we help organizations build a kind of institutional “metacortex.”
As patent creation becomes democratized, everyone becomes an inventor, everyone’s got a claim. Most of that innovation goes unnoticed or isn’t worth protecting. The real value isn’t generating more patents, but identifying ideas and knowing which ideas matter, when they matter, and how to protect them.
That’s why we started with deep prior art search and met engineers where they already work—Jira, Confluence, Slack, SharePoint. By grounding internal knowledge against the external world, we can identify what’s truly novel.
From there, we go further: mapping your IP against competitors, understanding who knew what and when, seeing how patents are used, predicting where others are headed, and surfacing innovation you didn’t even realize you had.
We built IP Copilot from day one knowing patent drafting would be democratized.
2026 is the year it happens.