When did we become an internal software company?

Hiya 👋 

There's a question showing up in a lot of conversations right now, in boardrooms, engineering standups, and this week, a pretty viral tweet. It goes something like: how much of this should we actually be building ourselves? This week's stories all orbit that question from different angles.

Let’s get into it 👇

A new piece from a16z asks a question that sounds almost rude until you've spent a week inside a Fortune 500: why do companies still run on SAP?

What's the story? SAP, Salesforce, ServiceNow - these platforms stick because decades of customization, approval logic, and institutional knowledge are baked into them. Moving off can cost $700M and three years. One German supermarket spent $500M trying to leave SAP and eventually gave up. These systems aren't sticky because they're good. They're sticky because they've become the encoded memory of how the business actually runs.

The piece's core argument: AI won't replace these systems. It'll wrap them. The interface becomes the new software frontier. Instead of your team clicking through 12 screens to update a vendor record or post an invoice, they describe what they need and the AI executes it through the existing system. The backend stays put. How people interact with it changes completely.

For mid-market operators, this reframe is practically useful. You're not trying to rip and replace your ERP or CRM. You're building the connective tissue between them — the layer your team has been bridging with spreadsheets for years. That's the part AI can now handle, at a cost that didn't exist three years ago.

If you run a small or mid-market company: Before you think about what to replace, map where your team spends the most time moving data between systems manually. That's not a technology gap. That's your starting point.

Go deeper:

📰 Read the a16z piece → Why the World Still Runs on SAP

The Wall Street Journal reported this week that OpenAI is cutting back on side projects and refocusing on coding and business users. The reason: Anthropic is beating them in enterprise, and doing everything at once caught up with them.

What's the story? OpenAI's CEO of applications told staff last week: "We cannot miss this moment because we are distracted by side quests." Last year the company launched a video generator, a social media app, a browser, and e-commerce features for ChatGPT. Anthropic kept its product list short, focused on enterprise and coding, and quietly took the lead. Boring won.

This is a lesson every operator knows but rarely applies to their own AI strategy. The companies doing best right now aren't running the most experiments. They're the ones who picked something specific, finished it, and made it a repeatable process. OpenAI is worth more than $300B and even they couldn't outrun the cost of divided attention.

The pattern here is one most mid-market companies are running in parallel: a portfolio of interesting AI idea meetings or pilots, no one accountable for seeing any of it through, and a year from now, nothing operationally embedded.

If you run a small or mid-market company: Count your active AI initiatives or “teams talking about action plans.” If it's more than two or three, ask which one is closest to actual, valuable action? Focus on that.

Go deeper:

📰 Read the WSJ story → OpenAI to Cut Back on Side Projects

Jason Staats, who covers AI for accounting firms, posted a no-frills guide to using Claude Cowork for real bookkeeping work. Forty thousand accountants watched it in seven days.

What's the story? The video isn't conceptual. It walks through an actual bookkeeping file, a 6,700-transaction reconciliation, receipt matching, and tax workpaper prep. No pitch deck, no "the future of finance" framing. Just: here's the file, here's what I did, here's what it cost, here's where I wouldn't trust it yet.

That number — 40,000 — is the signal. Accounting is one of the most process-heavy, document-heavy, compliance-sensitive professions there is. These aren't early adopters. They're practitioners who've been skeptical of AI hype for two years and are now watching colleagues do their actual work faster. When accountants move, it means the tools are ready and the workflow fit is real.

The "does it security" chapter at the 15-minute mark is worth watching on its own. It's the question every ops leader asks before letting AI anywhere near client data, and Staats gives a straight answer instead of dodging it.

If you run a small or mid-market company: Think about your most document-heavy, repetitive process. Not the interesting one. The tedious one someone hates doing every month. That's the one to run this on first. Accounting firms aren't special here, they just happened to have a good guide. The tech is getting better to be able to handle more. Try it out with your own team.

Go deeper:

🎥 Watch the guide → Claude Cowork for accounting firms (30 min)

Know someone who’d enjoy this newsletter?
Forward it along, or send us a note if there’s a topic you’d like us to unpack next.