--- title: "Turning Commerce into Code — Black 4" description: "Why AI is smart enough to write software and solve math proofs, but still sucks at the commercial stuff, and how to fix it." url: https://black4.ai/blog/turning-commerce-into-code/ date: 2026-10-26T00:00:00Z --- # Turning Commerce into Code Why AI is smart enough to write software and solve math proofs, but still sucks at the commercial stuff, and how to fix it. Joey SterlingOctober 26, 2026 Apollo's chief economist, Torsten Slok, just [put out a set of charts](https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector) that say what is obvious to many people implementing AI today. Look at the S&P 500 without the seven biggest tech companies, and revenue per worker is flat. Margins are flat. There's no sign of a productivity lift. His words: AI has delivered “zero financial payoff” to corporate America outside of tech. [TechCrunch picked it up](https://techcrunch.com/2026/10/25/ai-delivers-zero-financial-payoff-to-most-companies-according-to-apollo-economist/) the next day. We don't work with corporate America, but we see the same thing every day in smaller businesses. At best, leaders have gotten their teams subscriptions to powerful AI tools. (That's great. If you haven't, do it today.) But on its own, that gets you a team full of individual productivity gains that make people feel better and never show up in the P&L. You may also feel like you've armed your team with slop cannons (An AI tool that fires out plausible-looking posts, emails, decks and code faster than anyone can check them. Lots of output. Not much value.). AI is ready. It's been smart enough and cheap enough to work on your business for a while now. It can fix, or help you fix, almost any business problem you can help it understand. The problem is that the businesses themselves aren't ready. ## Code vs. commerce Point the latest AI at a software product and the results are pretty incredible. That's not because software engineers are better at prompting, or even because they're technical. A good engineer knows code about as well as a good marketer knows leads. It's because a codebase (All the code behind a piece of software, kept in one place with a history of every change.) is written down. The product does what the code says. Every change goes through version control, so there's a record of who changed what, when and why, and almost any change can be rolled back. Mature teams add automated tests that catch breakage, standards for how code gets written, a review before anything ships, and well-known technologies that every engineer, and every model, has seen a thousand times. Software has plenty of debt too. But even in a complex product, it's debt you can read. When an agent (AI that does the work itself, like updating records, writing and sending a draft for review, or fixing a page, instead of only answering questions.) asks how something works or what a change will break, the answer is in the code somewhere. We do this every day building the Black 4 platform, and so do software teams everywhere: many agents at once, across many parts of the product. They write the code, test it and open it for review, often working for hours without anyone stepping in. Now think about a commercial function: sales, marketing, support, any part of the business that isn't code. (That's most of it, at most businesses.) Point AI at your CRM (Customer relationship management: the system that keeps track of your leads, customers and deals, like HubSpot or Salesforce.) and it finds four definitions of a lead, a hundred fields nobody ever used, integrations with apps you stopped paying for, and plenty of stale, dirty data. Point it at your website and it finds tracking codes nobody can explain. It finds all the things we know are wrong with our businesses but never had time to fix. All of it is debt (The cost of every shortcut and leftover in your systems: old tools, old tracking, half-finished setups. It's not on the balance sheet, but you pay interest on it every time someone has to work around it.), built up over years. Think about the day you moved from Mailchimp to HubSpot. Or every time you switched SEO or ad agencies, and the new one added its tracking code and left the old one's behind. Pile that up, and even superintelligence would have a hard time telling you your conversion rate. We want to see commerce accelerate the way coding has. According to Anthropic's S-1 (The registration statement a company files with the SEC before it goes public. It has to tell investors how the business really works.), Claude wrote [more than 90% of the code](https://www.sec.gov/Archives/edgar/data/0002130357/000162828026067168/anthropicpbc-sx1.htm) added to Anthropic's codebase in September, and its engineers now contribute, on average, over 45 times as much code a day as they did in 2024. Imagine your marketing moving like that. ## Turning commerce into code Most businesses have very little to do with code, and their owners shouldn't have to learn it. But if you want to point a whole lot of intelligence at a business, the business has to work more like code. Written down. One version of the truth. Important changes tracked. The cool thing is that the work to get there, every brick that needs to get laid, now goes faster and better than ever. (Thanks, AI!) The agents we use to clean things up, and the platform we use to connect them, are built for exactly this. Cleanup that loomed for years, and never made the list because it cost too much time and money, now takes days or weeks. 1. Clear goals 2. Connected tools 3. Clean data 4. Rules & context 5. People & AI at work 6. The intelligent business ### 1\. Write down the goals What is the business actually trying to do this quarter? What are we optimizing for? It's probably the same goals you already put in front of your team, or keep in your own head. We write them down in one place, so the agents working for you are aiming at the same thing. ### 2\. Connect your tools Your CRM, your website, your ad accounts, your accounting: whatever systems hold the context of your business. We connect them so agents can read and work in the tools your business already runs on, and so the tools finally share what they know. Each connection gets the right permissions, so an agent can only see and change what it should. ### 3\. Clean up click to cash Click to cash (The whole path from a dollar spent on marketing to money in the bank: the click, the lead, the follow-up, the sale and the payment.) is the most basic question in commerce: we spent this dollar, someone clicked, they asked us something, we followed up, they bought, and the money came in. We clean up and define the data until that path is clear. One definition of a lead. One tool that owns it. One place where people and agents see the same numbers. Then we make it durable, so the numbers stay right as new data comes in. You don't have to clean up everything at once. Start where you want AI's help first. Maybe that's your support operation. Maybe it's running your ads. Get the data clear in that one area, and the agents can get to work there while the rest catches up. ### 4\. Write down what agents need to know What you sell. Who you sell it to. How you price it. What you promise, and what you never promise. What your brand sounds like. What years of experience have taught you. Slok argues that business runs on judgment, and that AI doesn't have it. Out of the box, he's right: no model arrives knowing any of this. So we write it down once, in one place, and every person and every agent works from the same playbook. That includes your brand kit, so everything the agents make looks and sounds like your business. ### 5\. Put everyone on the same layer Once commerce works like code, even in one area of your operation, people and agents can get to work on it together. You don't have to lock yourself into one AI provider. New tools come out every day, with different strengths and different costs, and each one is worth more to a business that has done the steps above. Sometimes it's you working on the business. Sometimes it's your team, an agency, or outside help you bring in. Everyone, people and agents, works from the same layer through the Black 4 MCP (Model Context Protocol: the open standard most major AI tools use to plug into other software. Connect a business once, and any agent that speaks it can work with that business, within the permissions you set.), and you still set the direction and approve what matters. This is the fun part. This is where you multiply the intelligence working on your business, with you, every day. ## Pay your debts Every business carries debt like this. For years, paying it off cost more than it was worth. Now the reward has never been higher, and the cost has never been lower. Untangle the integrations. Clean up the data. Turn commerce into code. Then bring in all the agents you want, and point them at every problem you've been putting off. That's an intelligent business, and that's where the payoff shows up. Explore what this could mean for your business. [Start a conversation](https://black4.ai/#contact)