The art of developing happy customers – ComputerWeekly.com

While the opinions of industry experts tend to vary, there is a growing consensus about what constitutes modern software development, and a few common themes emerged in the conversations Computer Weekly had when discussing the subject.

For Mark Holt, chief technology officer at Trainline, the history of software development has been about providing programming tools that offer higher and higher levels of abstraction. Increasingly powerful software is now just a download away. A database used to be a big scary thing with restricted access now you can download 15 databases from the internet, he says.

Bola Rotibi, a research director at CCS Insight, defines modern software development as the task of building cloud-native, cloud-first and multicloud applications. Its also about embracing data-driven big data insights and making use of artificial intelligence [AI] and machine learning, she says.

But modern software development is also about granular code reuse and low-coding tools.

Apart from everything cloud-related, there has been a shift in emphasis on the basic model that underpins a software-based system, to reflect the importance of data. AI represents the pinnacle of this data-driven model for application architectures. The AI is programmed by training it using sample data; it can then make decisions for itself using real-world data. The better the training data, the more likely the AI is to make the correct decision when presented with a dataset it has not encountered before.

Looking at the role of AI in software development, there has been plenty of talk on the web about OpenAIs GPT-3, a new AI for text processing. Its designers say it provides a general-purpose text in, text out interface, which can work on any English language task. In June 2020, OpenAI released an application programming interface (API) for GPT-3. In a blog post describing the algorithm, OpenAI said: Given any text prompt, the API will return a text completion, attempting to match the pattern you gave it. You can program it by showing it just a few examples of what youd like it to do; its success generally varies depending on how complex the task is. The API also allows you to hone performance on specific tasks by training on a dataset (small or large) of examples you provide, or by learning from human feedback provided by users or labelers.

In effect, programming GPT-3 involves showing it some examples; it then figures out everything else for itself.

Twilio software developer Miguel Grinberg recently uploaded an example of how he used GPT-3 and the Flack framework to build a Twilio chatbot using Python. What is intriguing about the application is that the steps he describes, which involve some pretty basic Python code, actually invoke one of the most powerful AI engines that exists, to provide human-like responses to random questions.

Why stop at producing English language responses? Some commentators on the web have used GPT-3 to write programs. Crowdbotics is one of the companies that sees GPT-3 as revolutionary for software development. In a recent blog post, the company wrote: We think the emergence of high-quality natural language interfaces will have a transformative impact on most tech tools used by humans. Any tech company with a product that contains user interfaces will need to come up with a strategy for how GPT-3 will affect their business or be supplanted by tools that use language more intelligently.

PayPal chief technology officer (CTO) Sri Shivananda believes AI could be trained to create some applications. Code is made of building blocks, which can be built on to make large, complex systems. As such, a programmer could write a script to speed up a repetitive task. This may be enhanced into a simple app. Eventually, it could become a payment system. At the very bottom of the software stack, there will be a database and an operating system. AI can assist in coding, says Shivananda. You can create knowledge in code and dynamic logic, but rule-based code writing can only go so far.

While some code will be written by AI, programming has differing levels of complexity, which means AI may be better suited to some tasks than others. Word processors, for instance, can correct sentences intelligently. Any document processor already offers a lot of basic grammar checking, says Shivananda. The word processor reads the sentences and applies the rules of grammar. The same technique is used in programming editors and interactive development environments to correct syntax.

Such rule-based checking has existed since programmers began using compilers and high-level programming languages to develop applications. Compilers and static code analysis tools effectively check that the lines of code are constructed in the right format.

But Boris Paskalev, CEO and co-founder of DeepCode, says: Compilers and code analysis are far from perfect. They are ultimately built and designed to catch and prevent specific issues that are part of the design of a given language, as well as pieces of the knowledge of the architects and developers building those systems.

Paskalev says non-trivial bugs and issues that exist in software development are related to the intricacy or ambiguity of the programming language used by the software developer. These nuances are covered by the code compilers or existing code analysis tools. In such cases, AI and the global development community come in by automatically learning from the hundreds of millions of bugs already solved by developers around the world and preventing/alerting every developer in case they are having the same or similar issues, says Paskalev.

He says DeepCodes system was recently used to identify an issue in a complex embedded system for engines, where specialists had been suffering for months searching for a serious problem. Paskalev says DeepCodes system, which is trained on millions of bugs that other developers have already fixed, was applied to the embedded system, and identified the problem in seconds.

Consumers have grown accustomed to seeing related products and recommendations when they shop online. The idea of having an e-commerce site suggest that people who bought a particular battery-operated toy also bought AA batteries, can and should be applied to programming.

The rule of never reinventing the wheel means software programming draws heavily on pre-built components, software libraries and, more recently, microservices, to enable programmers to add new functionality to their apps quickly. Thanks, in part, to the success of open source, programmers around the world are both consumers of open source code and are contributing code, by finding new ways to do things. Open source is a two-way street. You contribute and share like a social network, says Shivananda.

Programmers have always faced a learning curve to master a new programming library. But as systems have become more complex, the host of APIs, components and microservices available to achieve a given programming task has expanded beyond the capacity for any human to fully understand. Crowdbotics says GPT-3 can intelligently recommend open source code packages to solve development problems. Another possible use is to process a formal specification. As a programming aide, it may even be possible for an AI such as GPT-3 to check that the code a human programmer creates complies with the formal specification.

According to Paskalev, when using machine learning to capture all semantic logic and possible interactions, transitions and constructs in code, a well-trained AI will have no problem explaining what the programmer is trying to do. AI also offers programmers the potential to flatten the learning curve by drawing on the wisdom of the masses to infer what task the programmer is trying to achieve, and recommend the most popular approaches that others took.

We have already seen that working in a more localised/semantic way, but this will expand further towards business logic and larger architectural scope, says Paskalev. This expansion will require much larger datasets and computing power than we have today, as well as novel AI models that feed on the existing AI models and systems as data points. I call this an AI of AIs, encapsulating various AI techniques, representations, models and datasets. This is when we get closer to real AI and depart from the mostly augmented intelligence that we benefit from today.

AI will inevitably influence software development tools, supporting debugging and helping developers to write clean code quickly using the most appropriate programming libraries available on the internet. It is hard to predict whether GPT-3, or something like it, will replace hand coding, but low-code tools are gaining in popularity, because they lower the technical barrier to entry, so people from the business can write applications.

But for Trainlines Holt, although many of these tools are great for building a simple hello world style app that programmers often use to learn the basics of a new software development language, how many can work at enterprise scale?

The best line of code is the one you dont have to write, but there is a danger. You need to rein in using the latest thing, he says.

There are, for instance, numerous JavaScript libraries that come and go out of fashion. As Holt points out, the risk for an enterprise is that finding developers with the know-how will be much harder once the library is no longer fashionable. It is more important to focus on the end goal, and not necessarily get too enthralled in the tech to achieve it. Optimise your tooling, people and processes to create the best customer experience, he says.

Irrespective of the underlying technology, this is perhaps the main focus of a modern software developer modern software development is about building applications that deliver the best possible customer experience.

Go here to see the original:
The art of developing happy customers - ComputerWeekly.com