If your company is building some type of product or AI tool, congratulations! Now you are an AI company.
Yes, it is still a retail company. Or a bank. Or a CPG operation. You are that further an AI company: this is called a Ai too Company (AIAW): To advise a license to tell the sales and investor prospects that he is “doing AI”.
That license also puts it in the hook of new responsibilities. They are easy to omit at the beginning, but you will stop your true potential of AI if you do. And perhaps assume exposures to unnecessary risks in the process.
If Aiaws wants to make the most of AI, it would do well to take some hard lessons of the boom of software development technology. And in return, Software Dev also needs to learn some lessons about AI.
We have seen this movie before
Previously in my career I worked as a software developer. I quickly learned that any company that builds personalized software, regardless of its main business, had to learn the strings of managing a professional software products store.
Which was very good, except that they had no experience in execution of a software products store. The decisions of the executives were based on an understanding at the surface level of the personalized software, for the most part, “throw some developers into a room and tell them what to build”, which was enough to begin, but nowhere close enough to succeed.
If you forgive the “iceberg” analogy well used, most of what they needed to know about custom software existed under the flotation line. That’s where they would find things like “how to build a team.” (Do you remember the wrong jobs that required a computer title?) Then was “the need for separate development, QA and production time environments”, each of which requested their own hardware. That led to “we need to hire people to do quality control and manage operations.” The knowledge of the subsoil also included legal concerns such as intellectual property (IP), which fit with open source licenses … and so on.
That was a lot to learn. And yet, it was enough to get the initial product through the door, a considerable achievement, but that it is said that it executes only 20 percent of the total life cost of a software project. The time, effort and money required for long -term maintenance occurred as a triple sticker shock.
(The bonus lesson here is that the so -called “excessive” software that were trying to replace was not so expensive after all. But that is a story for another day).
There were also many strategic issues under the flotation line. Companies were not just With attention software for your business; That personalized software He changed how the business worked. The ability to execute certain processes 24/365 created new efficiencies and risks equally. The efficiencies were double -edged: the automation of a process could overwhelm the downstream processes that are still carried out by hand. The management of the new risks required that they all exercise a new discipline. A person forcing a change of hurried code could alter operations and lead to considerable losses.
These concerns are still maintained today, but they are mostly invisible, if not ridiculous, because software development has matured. The leadership of the company is well versed in the best practices of the industry. (In part, because many of today’s technological leaders are former developers who learned those best -hand best practices). But at that time progress was measured in hard -learning lessons, based on short knowledge horizons. Each step revealed more about the image of custom software, showing leaders that their previous understanding was simplified and not very fixed.
Some leaders retained experts in helping to protect their investment and accelerate their efforts. Others hindered on their own and finally discovered it. Or they did not discover it and suffered incidents of inactivity, high rotation and project failures.
We don’t have to relive the same movie
A similar story is developing in the space of AI. (For brevity, I will group all data science, automatic learning and Genai under the term “IA”). Like the development of personalized software of the first days, today’s opportunities have the price of new approaches and new discipline. You cannot simply put a group of data scientists in an office and cross your fingers so that everything works.
Many companies have tried. They have stumbled on the dark room that is AI, hitting their pimples and stepping on peaks because … I don’t know why. Hubris? Ego? A love for pain?
The newly coined AI today, as well as companies, such as its previous software counterparts, have to address the operational issues of this new technology. But before that, Aiaws must do the preparation work around the strategy: “What is Ai, really? What can you do in general and what can you do for us in particular? How can the incorporation of AI in our damage to us or our customers or non -affiliated parts that are simply in the wrong place at the wrong time?
Answering these higher level questions requires AI literacy, and that begins at the top of the org table. A leadership team that appreciates the complete scope of the abilities and weaknesses of AI is prepared to make realistic decisions and significant surface use cases. They know that they involve legal, public relations and risk management teams, early and often to limit the number of unpleasant surprises in the future.
And there are many surprises for everyone. The majority derives from the probabilistic nature of AI: models can exhibit a sudden increase in errors, either because they have reached a strange internal corner box or the outside world has changed. And that is if you can even make them work first. Like a financial investment, AI can give you 10 times the return or eat your money or anything intermediate. It can influence that result, but cannot control it: no amount of screams, cajoling or sessions everywhere can force a model to function well.
Then there are the new risks that AI brings to the table. The models will inevitably be wrong from time to time; How do you handle that? How often can they be wrong before being in hot water? Do you have a license to use this training data for this specific commercial purpose? Is that model allowed to operate in each jurisdiction where it interacts with its end users?
Wait that some of those legal questions are in flow for a while. You can win sitting in the gray area of regulatory arbitration, but only if you are prepared for a fast pivot when those limits change. And that is just the Court of Justice. You also face the public opinion court. The practices of AI that are considered spooky or invasive can trigger a public reaction. (Suggestion: You may want to avoid facial recognition for now).
You will notice how much land I have covered before talking about hiring. Taking AI to a company means that it has new roles to fill (data scientist, ML engineer), as well as new knowledge to fill in existing roles (products, PAHO). Companies that begin their journey from hiring data scientists are skipping a lot of preparation work, at their risk.
Running the list of lessons for Aiaws, there is surveillance. AI is a changing landscape. There is no viable approach to “establish and forget it.” The roles, strategy and execution require periodic review and adjustment.
A strong weak point
The AIAW that execute strong software development stores are, contraintuitively, ready to learn these lessons in the difficult way.
That force of the software is doubled as its weakness of AI. Since the development of applications and AI imply the writing code, overestimate the overlap between the two. We know Python. All these things of ia is python. How difficult could it be?
These companies adopt AI in the same way that some developers move to a new programming language: hold on to the mentality of the old. Its code can pass to the Python interpreter, but they are all Java’s constructions. Java Python is difficult to endure and does not take full advantage of what Pythonic Python has to offer.
So what is AI with software development? It is the CEO who assumes that, when using a popular API LLM or another product of Ai-Asvice (AIAA), they will not need any experience of internal. It is the leader of the product who announces characteristics backed by AI before the models have been demonstrated. Or expect the consistency of software behavior once the models are in service. It is the one who is so dead so that his efforts of AI fit agile that never look for the specific best practices of AI. This person is the main developer, who believes that his model is ready for the star schedule because he has followed the tensorflow tutorial.
In general, it is the company that advances in the AI at high speed, driven by a self -confidence that exceeds its knowledge horizon. This arrogance injects unnecessary frustration and risk exposure in their efforts.
The funny part is that this crew could obtain a product of AI by the door. But they will not realize the hard truth: just because racing It does not mean works.
At least, they will not realize until after the application enabled for the AI interacts with customers and promotes commercial processes. The inevitable problems will be much more difficult to address at that time.
The future is the past again
In fact, I drew the first part of this article several years ago. At that time, he focused on companies that enter personalized software. It was disturbing to meet the old scheme and see the same story playing in the world of today.
Ironically, the AIAW that are better to create software to learn more. First they need to unlearn certain software practices to create AI solutions using an AI mentality. But once these companies really adopt the best practices of AI, they are also positioned for the greatest victories. They already understand ideas such as activity time, implementation, version control and monitoring, which is all that is required once the R&D model and production move.
On May 8, O’Reilly Media will stay Coding with AI: the end of software development as we know it—An Live Virtual Technology Conference that highlights how AI is already overlapping developers, promoting productivity and providing real value to their organizations. If you are in the trenches that build tomorrow’s development practices today and are interested in talking in the event, we would love to know about you before March 12. You can find more information and our call for presentations here. Do you just want to attend? Register for free here.
#Congratulations #ORilly #company
