OTEL Logging: The General Ledger for AI

If you want to implement OTEL logging at your organization, there are plenty of how-to articles on the internet — great content that I don’t want to recreate here. Instead I want to explore what OTEL logging unlocks over the next 12 months.

The historical analogy that comes to mind is double-entry bookkeeping, a boring invention that had an outsized impact. Just keeping better track of money in and out, what you own and what you owe, increased people’s confidence to allocate capital and expanded economic output.

Good logging for AI is a similar invention. It’s a dataset that creates confidence, and moves us from vibes to clear(er) thinking. Here are a few predictions on where this is headed.

1. The Intelligence Manager

For five hundred years, finance has managed the scarce resource at the heart of the economy: capital. The new scarce resource is AI tokens, and managing it might look like accounting.

Expect a new role to emerge, the intelligence manager. An intelligence manager does a wide range of tasks with a single-minded goal of getting the most out of AI spend.

When it comes to logging, this person understands what workflows AI is being used for, the per-workflow cost, how many tokens are being spent productively (or not). The intelligence manager understands where the organization can get the most life through training employees or switching to different AI products.

The intelligence manager might live outside the IT org, might report directly to the CFO. My prediction is that by the middle of 2027 AI Operations job listings are as common as DevOps listings were by 2018.

2. AI Security Grows Up

For two years, AI security has been security theater — a lot of “Microsoft Copilot is the only safe thing” and fear-driven messaging. Between security professionals and the rest of the world we didn’t have a good shared view of what was really happening, and so fear stepped in. Good logging hopefully ends that.

Once you have the traces, AI policy stops being a document no one reads and starts being a system you can audit, enforce, and defend. I hope this is the moment IT departments have been waiting for, to grow up and get more involved in solving the problem.

My prediction is by the middle of next year the relationship between security teams and the rest of the business will be much healthier. IT and security has been the department that says no to AI. With telemetry, there’s the potential for IT/security to be an enabler. The technical professionals who make the shift are going to be heroes.

3. The First Real ROI Numbers

For most knowledge work, we measure cost (salaries, hours) directly, outcomes indirectly (revenue, profit), and the processes themselves hardly at all.

Because we can observe AI-assisted computer work directly, AI logging presents an opportunity to see how knowledge work gets done, to allocate time and costs at a task level.

We’ve never had a unit price for “do an SEO audit” or “build a quarterly board deck,” but now those tasks have a measurable cost in dollars and seconds. It’s time and motion studies for knowledge work that are like exhaust from the systems.

My expectation is that we’re going to discover something incredible: across a wide band of knowledge work, AI will be very competitive with human labor on both cost and quality, and that will drive large increases in spending on AI.

My prediction for the middle of next year is the first wave of internal “AI cost cards” — literal price lists for common deliverables, AI vs. human, with quality scores attached.

4. Open Models and the AI Workstation

With greater visibility into AI outcomes/costs, more confidence in the security profile, and a full-time person managing the ledger, expect firms to be more aggressive in driving down costs (as they spend 10–50x more).

I think models like Opus 4.7 / GPT 5.5 are an early version of AGI and we’ll have open source models of this capability in 12 months. Good logging should make it easier to evaluate these closed source models against new open source alternatives.

At the same time that AGI open source models are coming online, roughly half of the hyperscale data center projects might be paused or stalled. The answer might be to decentralize data centers.

Every knowledge worker is going to want an always-on computer to run agents and it’s going to be a lot easier to spend $20,000 on an AI workstation for your best knowledge workers.

The Mac mini / openclaw craze was an early signal. What does that look like when compute is constrained and we want to run frontier-size open-source models locally?

My prediction is that early next year we’ll have a clearly-marketed (and popular) “AI workstation” SKU from Apple or Dell in the $20,000+ range.

The AI Ledger Increases Confidence

Double-entry bookkeeping didn’t make merchants rich. Good accounting lowered transaction costs for the merchants who adopted it, and that led to an increase in economic activity, at the firm level and for the whole economy.

The same dynamic is starting now with AI logging. The organizations that build an AI ledger this year will be the ones making more rational decisions about hiring, security, ROI, spend, infrastructure. Everyone else will still be vibing.