Hey, I’ve Been Thinking…

Weekly #5: The Week the Interface Started to Disappear

Written by

Adem Ogunc

in

April 19, 2026. A weekly technology roundup written by a founder who builds with these tools, not just reads about them.

This is my fifth Sunday writing one of these. Every week I sit down with a coffee, go through what I read, built, and played with over the past seven days, and try to make sense of it for anyone who wants to follow along. You dont need to be technical. You dont need to be in the industry. You just have to be curious.

This was one of those weeks where if you only looked at the headlines, you would miss the real story. The real story is that a bunch of things happened at once that all point in the same direction. The interface is starting to disappear. Salesforce is going chat first. Claude Design shipped. Robotaxis expanded in a way that was half real and half theater. Netflix posted its best quarter ever and the stock still fell. And somewhere inside all of this, if you squint, theres one architectural shift that ties it together.

Let me walk through it.

I tried to draw a yarn ball with Claude Design

Anthropic shipped Claude Design this week, and I spent four or five hours in it building front ends and wireframes. To be fair, Opus has been able to produce design mockups for a while now. React components, JavaScript, HTML. Its not a net new capability for me.

Whats different is the framework. Its built for the web, which sounds obvious until you are actually in there and you realize how much faster the iteration feels when the tool is designed for the output format. Structural UI work. Dashboards. Layouts. Components. All of that flows.

I tried to draw our original Well Woven logo. Its this yarn ball that unrolls into the wordmark. Icons? Nailed it. Descriptive illustrations? Not quite there yet. It gave me back something that looked like either a 2003 dot com logo or a 2012 Etsy shop. Neither is the vibe.

So the imagination heavy drawing side is not there yet. But for anything structural, its legitimately useful. If you have been meaning to try it, this is the week.

The Claude Code app experience might be the real unlock

I also started using the new Claude Code experience on the app this week. This might actually be my real productivity unlock of the last seven days.

It has an auto mode where it keeps working recursively without asking permission every two seconds. It integrates with CodeRabbit so it can critique and correct its own work in loops. Running Claude Code inside the app feels meaningfully different than running it in a terminal outside. Longer working sessions. Less hand holding. More actual forward motion.

I havent tried Codex yet. I keep hearing its better than Opus on long coding tasks. Thats the honest thing about this space right now. You cant use everything. The cost of staying current isnt just subscription fees, its cognitive overhead. For now Im in the Anthropic lane, getting results, shipping.

Salesforce went chat first and nobody is talking about it enough

This is the story of the week for me.

Salesforce announced this week that every one of their tools will be accessible through a single chat interface. Think about what that actually means.

Salesforce. The company that trained a whole generation of business people to click through 47 tabs just to update a single opportunity. Theyre now conceding that the primary way you interact with their product is going to be conversation, not clicks.

Thats not a feature announcement. Thats a category bet. And its the clearest signal yet that the enterprise software world is going headless.

What headless actually means

Headless gets thrown around like jargon so let me ground it.

Traditionally software has two parts stapled together. The engineering underneath, which is the database, the business logic, the integrations. And the interface on top, which is the screens, the buttons, the dashboards you stare at all day.

Headless decouples them. You keep the engineering. You strip away the fixed interface. Anything, including an LLM, can query the engineering through whatever interface makes sense in the moment.

Picture a car. For decades you bought a car and the body and engine came as one package. You liked the PT Cruiser shape? Great, you also got the PT Cruiser ride. You wanted German engineering? You took the German body with it. Headless is saying keep the engineering platform you want, and snap whatever body on top suits the job. BMW engine. Jeep body. Whatever gets you where you are going.

Or think about it the way Ive been thinking about it. The Mac versus PC thing. Weve all had this argument. You prefer Finder, I prefer Explorer. Your folder icons look one way, mine look another. But when you actually sit down to work, you are doing the same tasks. You are writing a document. You are building a spreadsheet. You are sending an email. The operating system preferences are real, but they are texture on top of identical outcomes.

What happens when the LLM becomes the universal interface? The OS starts to recede. It doesnt disappear. You still need something to run the thing on. But it stops being where the design argument happens. The design argument moves to the agent layer. To the workflow. To what happens in the background while you are focused on the actual work.

Why this matters if you actually run a business

If you run a business, you live in spreadsheets. I live in spreadsheets.

Every Monday Im pulling Amazon advertising data, Shopify sales, Wayfair orders, inventory positions across 10 sales channels, ranking data, FedEx billing disputes, returns. All of it into Excel.

Excel isnt where the data lives. Excel is where I force the data to converge because I need one interface to see all of it at once.

In a headless world I dont need Excel as that convergence layer. I ask a question in plain English. The agents, or services, or whatever we are calling the worker bees this week, go query the systems. The answer comes back. No more tab switching. No more VLOOKUPs. No more reformatting a Wayfair export so it matches my Amazon SKUs. No more Monday morning spreadsheet choreography.

Thats the promise. And Salesforce going chat first is the most expensive, most enterprise level, most “this is actually happening” version of that promise I have seen so far.

If Salesforce is going headless, every enterprise SaaS vendor is going to have to answer the same question within 18 months. Whats our chat native story. What happens to our UI moat when the UI isnt the moat anymore.

The operators paying attention to this right now will have an absurd advantage in 2027. The ones still buying software based on which dashboard they like best are going to wake up lapped.

Robotaxis had a big week. Watch the units shipped.

Tesla launched unsupervised driverless service in Dallas and Houston on April 18. Waymo opened Miami and Orlando to 150,000 waitlisted users on April 15 and started public road testing in London on April 14.

Big headlines. But heres the operator detail that matters to me.

Tesla launched Dallas with one active vehicle. Houston with one active vehicle. Thats not a product launch. Thats a press release with a permit attached.

As someone who measures scale in units shipped, not announcements, I notice when the gap between what gets announced and what actually gets deployed is that wide. Waymo, to their credit, onboarded 150,000 waitlisted users into Miami and Orlando. Thats a real launch. Tesla is doing theater.

Both companies are in the same news cycle. They are not doing the same thing.

Netflix posted a blowout quarter and the stock dropped 10%

Revenue up 16% to $12.25 billion. EPS of $1.23 against $0.79 consensus. Paid members over 325 million. And the market punished them because Q2 operating margins are projected to dip 1.5 points, and because Reed Hastings is leaving the board in June.

Heres the operator takeaway. The market is done rewarding growth at any margin, even from the poster child of subscription businesses.

If Netflix cant get a pass on a 1.5 point margin dip, nobody can. For DTC founders, this is the last reminder you need that margin discipline isnt a 2023 concern. Its the only concern.

OpenAI and Anthropic are bifurcating into specialized models

OpenAI dropped GPT-5.4-Cyber for security researchers on April 14, and GPT-Rosalind for life sciences on April 16. Anthropic had already announced Claude Mythos under “Project Glasswing” earlier in the month.

The pattern. The frontier labs are splitting into general purpose consumer models and vertical specialized models.

If you are a generalist consumer using ChatGPT or Claude for everyday tasks, this does not change much for you. If you are an operator trying to get a real workflow done, the interesting stuff is moving to the verticals. Watch this space. The next big productivity unlock for e commerce probably isnt a smarter general purpose model. Its a retail specialized one.

The take I will get yelled at for

Everyone in my feed right now is obsessing over agents. Agents, agents, agents. Build an agent for this. Build an agent for that. Agentic this. Agentic that.

I think the agent conversation is a distraction from the actual architectural shift, which is the decoupling underneath.

Agents are a consequence of headless, not a cause. Build the headless foundation first. Clean APIs. Queryable data. Documented integrations. And then the agents follow naturally.

Try to build agents on top of a stack thats still locked behind proprietary dashboards and you are going to spend the next two years wiring bailing wire around the same old SaaS tools. That isnt an agent strategy. Thats cosplay.

The unlock is the decoupling. The agents are downstream.

If you are a small or mid sized operator, the most valuable thing you can do in Q2 isnt build an agent. Its audit where your data actually lives and how queryable it is from outside the system.

Can you query your inventory from a script, or only through the UI. Can your CRM data be pulled in natural language, or do you have to export CSVs and reshape them every time. What systems are you paying for right now where the data is effectively trapped behind the dashboard.

Thats the audit. Thats the work. Thats what sets you up to ride headless when it becomes the default instead of the exception.

Community is where the learning is. And where it isnt.

Final thought.

I keep seeing people post screenshots of everything they are building, sharing prompts, comparing tool outputs. Thats the good version of the internet right now. The bad version is the guys yelling the loudest on X about what is coming. You dont learn anything from the guys yelling. You learn from the people actually shipping.

If you havent tried Claude yet, drop a spreadsheet in there. Drop a deck. Ask it something you would normally Google. Be persistent with it. The technology is a hundred times more capable than most people realize, because most people bounce off after one lukewarm response and never come back.

Stop consuming content about this stuff. Build something with it. Thats where the knowledge actually lives. Always has.

See you next week.

This is Issue #5 of “The Week in Technology,” a weekly newsletter at ademogunc.com.

Sources and Recommended Reading:

About me: Im Adem Ogunc. I run Well Woven, a rug company based in Easton, PA, and FurniPulse, a home furnishings trade intelligence platform. Ive been in the rug industry for over 20 years and building with AI tools for the last couple of years, using them to run advertising, manage operations, and build our e commerce infrastructure. I write this newsletter because Im genuinely fascinated by whats happening in technology right now and I wanted a place to think out loud about it. If youre curious about any of this, whether youre in tech, e commerce, home furnishings, or just trying to figure out whats going on, Im glad youre here.

Issue #5. See you next week.

Comments

Leave a comment

More posts