Artificial intelligence cannot accelerate software delivery.

During the years of my growth, I had a Labrador-Whippet mix named Barclay. We were almost inseparable. When I was digging holes in the garden, it would sniff around beside me. When I was reading a book, it would lie on my lap, often making my legs go numb. When our whole family went for a walk in the forest, it would run ahead to scout the way and then turn back to check on us. As it ran back and forth merrily, the distance it covered was almost ten times that of ours.

The best thing about dogs is that when you're going through a tough time, you can confide in them. They are excellent listeners and understand you when no one else does.

If I were to get another dog, companionship would be my top reason. There are many reasons to own a dog, but for me, companionship always comes first.

Now, imagine that I go to a rescue center to look at adoptable dogs, and they want me to take a cheetah home just because it runs much faster than a dog. This is like some organizations claiming to be using artificial intelligence but only narrowly focusing on speed.

Speed is never the goal

We've been through this before. We should all feel sorry that the agile movement has withered into a hollow shell, leaving only the simple pursuit of "speed." Speed has never been the ultimate goal. The primary significance of improving work efficiency is to get feedback earlier. When you find that your amazing new feature doesn't interest users, you can stop developing it immediately.

If you can terminate a bad idea early, you'll waste fewer resources and can immediately switch to a better one. No one should blindly pursue developing software with the most features and the fastest change speed. When software is piled up with too many features and the ever-expanding feature list is frequently changed, people will start to dislike the product you're developing.

Microsoft Word is the most powerful and feature-rich word processing software today, but few people use it specifically. They're all using Google Docs, which has far fewer features. This shows that the features Google chose must be more attractive, or that fewer features make the software more user-friendly. In fact, it's the result of many small factors working together. Sometimes, one truly outstanding feature can overshadow all the others, and the convenience of being able to collaboratively edit Google Docs in the browser might just be such a feature.

If you had asked people twenty years ago, they would have told you that Microsoft Word had an unshakable position among similar software, but now it only has a 3.9% market share, while Google Docs has 9.6% (Source: https://6sense.com/tech/productivity/googledocs-vs-microsoftword). If you think this market shift is just a pricing issue, you're probably working for an organization that only one-sidedly pursues speed, because you no longer believe in the idea of building software that users truly value.

Adopting AI for speed lacks credibility

Software industry leaders have their usual ways. When they claim to adopt artificial intelligence just to blindly pursue speed, this should be viewed with caution. You'll find that in the past ten or twenty years, they've announced agile transformations, adopted DevOps, or carried out platform transformations in pursuit of speed.

They've spent a lot of energy on all these initiatives without achieving significant results, which fully shows that they're not as eager for speed as they claim. Of course, they want to put a "DORA Elite Performance" badge on their work resumes. But they still don't have the real motivation to improve speed because they're not interested in the basic result of more frequent deliveries - user feedback.

Any leader who has put their team through so many troubles in the name of speed and now says that artificial intelligence will finally bring speed is deceiving themselves.

Feedback metronome

When you value feedback more than speed, you'll let the feedback loop drive the rhythm of the entire software delivery process. Setting the rhythm with this beat will give you the crucial leeway to handle feedback and achieve the core advocated by the agile concept: quickly adjust the direction.

Organizations and teams that use feedback as a metronome to set the overall work rhythm often take the initiative to find and eliminate all work that disrupts the rhythm. They plan the team structure to work with the smallest and carefully designed dependencies. They simplify the change approval process to ensure that the team can independently decide when to deploy and can also fully observe the actual effects after deployment.

The DORA model and its included generative culture, transformational leadership, lean product management, and continuous delivery process were not created by chance but are the results of decades of in-depth practice. Teams that practice these concepts do have delivery speed, but this is not the original intention of their adoption of such culture and practice. What they really pursue is to obtain high-frequency and high-quality feedback to figure out what really needs to be developed.

What the Elite team does

The Elite team is a software team under a large healthcare enterprise. The organization mainly engages in software business related to patient management and emergency triage. In the safety-critical industry, this kind of software can really be a matter of life and death.

They release the patient management system every six months, and the test cycle of the decision support system is two weeks. If problems are found, two more weeks are added. This cycle repeats until a version passes the test!

Even with such a past, we still managed to carry out a six-month work plan, creating a deployable software version every three hours. We followed a very powerful set of technical practices, but what we cut out might be more important than what we added. This balance comes from introducing executable specifications for instantiated requirements and eliminating the slow and inefficient bureaucratic review process.

In terms of results, we reached an important cooperation with a new healthcare service provider. They needed to provide a decision management API to connect to their website system. We safely delivered a usable API within two weeks and facilitated a contract worth $1.8 million (equivalent to $2.5 million now).

If your organization hasn't sorted out the production path and made similar changes, there's no way to achieve the delivery speed you want. You'll introduce artificial intelligence just like you introduced Scrum, DevOps, and platform engineering before, and still end up with little effect, just like in the past.

The most important thing you can do now is to sort out the value stream, especially the process from code submission to production deployment, and start fixing the broken parts. There's no mystery about what changes need to be made. Dave Farley and Jez Humble revealed the secrets in their co-authored book "Continuous Delivery: Reliable Software Releases through Build, Test, and Deployment Automation."

Why do you adopt artificial intelligence?

Before this, you might have said that you adopted artificial intelligence to improve efficiency and speed up progress. If you take software development seriously, I hope you can change this statement and clearly tell those around you that frequent feedback and decision-making agility are one of your core priorities.

Teams that have solved the throughput/stability trade-off through practices such as continuous delivery have a lower obsession with simply pursuing speed. They're more willing to explore more valuable development opportunities. Finally, I'd like to list and recommend several such opportunities.

Smaller teams are better. We made a compromise on this issue because what we needed couldn't wait for a very small team to complete slowly. We might double the team size, even though we know it won't cut the required time in half. The COCOMO model has a precise calculation of this law of diminishing returns, but Fred Brooks put it more impressively: Adding people to a delayed project will only make it more delayed.

Therefore, most software teams that understand the complexity of communication and coordination keep the number of people in the range of 6 to 12, the so-called "two-pizza" teams. But this is not the ideal team size; it's actually still too large. It's just a practical choice after weighing multiple factors. Now that we have artificial intelligence, we should consider building "one-pizza" teams, or even smaller.

Small teams with a high degree of autonomy and working based on loosely coupled components might be a powerful way to unleash the value of artificial intelligence in software development.

My last suggestion is that instead of just developing the existing software faster, artificial intelligence might enable your team to create products with a greater vision. You can start promoting the global business layout that you've been hesitant to start. You might have had some feature ideas but never been able to sort out a clear idea. With the help of artificial intelligence to quickly build prototypes, you can conduct explorations and practices that were difficult to try before.

In any case, start by optimizing the software delivery process and the deployment pipeline more vigorously. Make sure to close the feedback loop and use it to control the overall rhythm like a metronome. After doing this, when you introduce artificial intelligence, you can pursue goals that are far more valuable and imaginative than "simply increasing speed."

Original link: https://thenewstack.io/feedback-driven-ai-adoption/

This article is from the WeChat official account "AI Frontline." Author: Steve Fenton; Translator: Mingzhishan; Planner: Tina. Republished by 36Kr with permission.