The AI Bill Is the Cheap Part

Does it feel like your team is experimenting nonstop and working hard, yet you’re not really moving faster? Is the board scrutinizing your AI costs ballooning like my expenses when I enter an Italian menswear store?

Startups with 30+ engineers easily waste more than $1m a year on neglecting how their engineers operate. Isn’t it ridiculous to put all this focus on the AI agents? The real difference lies in helping your people. AI only multiplies whatever you already have in your hands. A poorly functioning team will just go about its way, producing poor results ever faster (and sometimes, ever poorer).

The Wrong Optimization

Turning your attention to tokens and compute cost is myopic and lazy. We’re reaching for the easiest thing right in front of us, neglecting the real issues. I remember one of the first rules I was taught in my first professional coding role: you don’t optimize without measuring first, or you’ll end up optimizing the wrong problems.

So let’s look at the measurements: what really afflicts tech orgs? And we don’t have to limit ourselves to the companies I happened to work with; there’s a lot more data available. Atlassian and other surveys routinely show more than 15-20% of engineers’ time getting wasted in bad processes and quality issues.

Even worse, I find the issue isn’t time clearly wasted, but the time we don’t realize is wasted. Beyond firefighting, there’s misalignment, misunderstandings, and busywork. Tackling those would result in a much bigger bang for your buck as a leader.

What I’d Prioritize Over Tokens

My recommendation is to collect real evidence of what’s happening across your organization for a period of two weeks and see where your time and costs are really going. To help you know what you’re looking for, here are the most common issues I tackle with clients, along with tips for handling them.

Firefighting

Quality issues, supporting the existing codebase, and responding to urgent requests often leave teams thrashing without making substantial progress on their declared priorities. How much time each week or iteration ends up going into unplanned work? That’s a good metric for time going to waste, which you can then trace back to the root causes and tackle them:

• Problematic code that needs to be redone? Migrations and rewrites now can be completed in days when they used to take months and years. If it’s sucking up hours every week, focused treatment could pay off quickly.
• Lots of quality issues? Is it due to scaling problems, poor use of AI and lacking reviews and testing, or too much rushing? Find the root causes of issues, determine whether it stems from older code or newly deployed features, and put in place the right protections to find it on time and avoid it.
• Incoming urgent requests? Stop treating each question and wish as a P0 issue. Clearly document and communicate the impact of these rapid changes to your peers in the executive team to realign on business priorities.

Misalignment

This is about work that’s just not important. For example, when the engineers tackle edge cases that don’t matter at the smaller scale. On the other end of the spectrum, we see Product and Engineering deciding to prioritize work that’s not really the most important and impactful due to various reasons. For example, you might already have a “Product X” team and thus keep feeding it work, even though Product X might not be the priority this quarter and the team might be better used helping out with other things.

Ensure that the team has enough context about everything it does to make better decisions, and make it clear that descoping features with Product before starting to work on them is important not just to save on tokens, but also to save our own precious time and focus.

Misunderstanding

When what was done isn’t what the other party meant. These happen more and more nowadays because product specs seem to be more and more haphazardly put together. Overwhelmed Product teams find themselves pressed for time and generating prompt-based specs that turn short ideas into detailed instructions without really understanding the issues and how all the different pieces should fit together in the end.

The context is important for tackling this issue as well. Engineers should have enough product mastery and understanding of the work they’re being asked to perform to be able to sniff out when things aren’t clear to them.

Work should be performed in smaller iterations to verification. That’s going away because people no longer have the healthy friction of needing to type the code themselves. Due to that, many PRs are too big and far ahead by the time someone else sees them. It’s still a bad idea.

Internal Overhead

We all like complaining about too many meetings and similar. Of course that could be the culprit, but there are also other internal issues, like having teams become too small, resulting in too much communication and management overhead. Often these sorts of issues are more easily spotted by someone external to the organization.

Taking the time to ensure that your organization and its operating system are streamlined will save countless hours and cognitive load that is worth much more than another Claude subscription or two.

Ossification

Lastly, there’s the issue of doing things in a way that used to make sense in the past but no longer does. When time gets eaten up in all of these previous issues, teams ossify, missing opportunities to evolve and improve. They keep trying to work harder to solve something that requires a leap forward and finding a smarter solution.

Injecting habitual innovation is the only sustainable path forward, and lucky for us it’s just gotten that much easier with AI. You can learn more about this by getting the free sample chapter of The Tech Executive Operating System that’s exactly about this topic.

Don’t Get Distracted

Yes, the token talk is all the rage. You’ll get lots more clicks on LinkedIn talking about how you lowered your AI costs by 30% than working hard on reclaiming 5%, 10%, 20%, and sometimes even 50% of the team’s time. Nevertheless, it’s only the latter that would unleash your team.

Fix the root causes and work better. AI is usually the cheapest line item on your budget, but it happens to have the most eye-catching invoice.