An idled factory (Gemini 3.8 Flash) I must admit, I was a little stumped at first. Growing up in the data science communities circa 2018, it seemed like whatever we solved with machine learning at the time only presented more opportunities. In many cases, successful automation grew the backlog instead of reducing it.
Is AI actually burning down the backlog? My answer: kind of.
This is the first question I ask. In many cases, the answer is that companies using AI are iterating faster over code, but other parts of the SDLC are under significant pressure. As example, AI can ship code that’s PR ready orders of magnitude faster than humans. But that puts pressure on testing and CI/CD systems that can’t keep up with the volume and pace. This is also the next easiest area for us to focus on because, at its core, how we approach the work is similar to code generation. We can create tests with AI, and even use AI to split apart CI/CD processes so we can increase the overall system throughput.
Now you could probably comment me any one company that’s breaking this paradigm, ships quickly, and has AI integrated throughout the SDLC. But I’d probably tell you this is like learning about Walmart in undergrad. There’s always an example, and then there’s the big middle.
So, we can use AI for more technical work, and you can see a number of ideas already moving in this direction. But it’s also breaking more than fixing in a number of organizations. Which leads me to ….
Does AI 10x good software engineering? My answer: absolutely.
Companies that have good engineering rigor are seeing more benefit from AI because they’ve done the dirty work by building microservices, decoupling key business areas with contracts and SLOs, and having clear developer processes. An easy example is a monolith. These require long build times for CI, and agents can stack up code to test 10x faster than any human can.
So you’re left with three options, you can buy a lot more compute to run the tests (not easy when you’re already dealing with tokenomics), you slow down the flow, batching up changes (which reduces your overall value from AI and increases shipping complexity), or you refactor what you’ve done, ideally with AI (what I find myself recommending more and more). Teams that invest in the infrastructure improvements will see higher returns. AI is a 200 MPH (~320kph) sports car that goes fast on well built roads. If your pipeline forces offroading, and slow going, AI will compound a wreck.
For many companies, it will take time but they are moving in the direction to refactor based on the technical gains you get, and that AI is good at refactoring already, and will get even better quickly. Which leads to my last question.
Will the backlog run out? My answer: I used to say no, but now I’m not sure.
In most discussions I have with leaders, I frame AI as ultimately a business transformation problem, not a technical one. I haven’t received push back yet. But what that also means is that we’re doing all the technical work today while we still try to figure out what it means for any company to build AI .
My general assessment of most AI implementations is that, outside of developer experience ones, we’re still very early, or dabbling. And, 4 years in with LLMs, I think I would have expected to see more movement. I’m not sure personally if this is due to my change in expectations of the rate at which we can do work or how fast the world moves, but true, agentic implementations still seem relatively sparse.
This suggests to me that we are still working out the business problem of AI. How does this change interactions with the customer? Does the customer care? What would we measure success by? These are all valid questions but they challenge the core business models companies are based on. They are uncomfortable discussions, and many times we don’t want to talk about it as a result.
But it feeds uncertainty to the backlog. Eventually features will get shipped faster, technical debt will start being automated in resolution, and then the business questions will come back, only in a much more urgent way.
I have 3 chips, where am I placing them? How do we force the business question in a productive way? Ultimately, the company exists to minimize economic friction, and to create value.
(1 chip): Business Model Generation and its companion Value Proposition Design are two books I highly recommend. These books force you to take a look at your service lines and products from the lens of your customer. I hope it creates discomfort, but in a productive way.
(1 chip): Throw your best customer and AI talent at the problem. I don’t think everything here will stick, but when we’re all trying to grapple and figure it out, those that figure it out the fastest win. This (in my opinion) is somewhat why you see Forward Deployed Engineering (FDE) roles across so many companies. They are embedded in customer organizations trying to help them use the product. The faster you get signal back, the faster you get first mover advantage (and new backlog) as a result.
(1 chip): Intentionally shred the org chart. In a disruptor model, the small and nimble usually win (evolutionarily this is true too). Create incubators in your company where the constraints are minimal, and the risk / reward scale is highly weighted toward rewards.
Ultimately, these activities are designed to find the new value opportunities, and then get the backlog back. As we build muscle around AI, the problem of ‘what’s next’ will take care of itself, but we need to start investing in the front of the pipeline to see the payoff.
So, what happens when the backlog runs out? The backlog question ultimately comes back to the value proposition of the company. Companies that cease operations don’t have a backlog, but it’s not usually the code that was written; it was because someone didn’t want what they were selling.
For companies, focusing on building talent that’s expert at the user and the user domain is the top skill. Invest in user centricity.
For the individual, keep investing in your own skills in AI to make yourself transferrable and marketable. Ultimately, investing in your own transformation will help your (and your future) companies as well.
