Artificial intelligence is able to write code, build websites, and help developers with jobs that once took hours. And that’s a reasonable question for anyone investing in coding education: If AI is getting so good, will kids still need to learn these skills?

That’s a particularly important question for entrepreneurs considering a STEM education franchise.

After all, technology is changing fast. An education business has to sell more than lessons built around tools that could look very different in five years.

But there’s something interesting going on underneath the headlines of AI replacing programmers.

In a 2026 study, researchers from Harvard University found that AI coding agents helped engineering teams produce significantly more code, but those increases did not translate into comparable gains in completed software output.

The reason why? Writing code is just part of creating something that actually works.

That distinction changes how we should think about the future of STEM education for students, educators and would-be franchise owners.

What Harvard’s 2026 Research Says About Coding AI

Researchers at Harvard University, Fiona Chen and James Stratton, released a working paper in August 2026 titled “Artificial Intelligence in the Firm: Bottlenecks in Software Production.”

The researchers studied around 300 million engineering work events from 718 companies to understand what transpired after companies adopted AI coding assistants and agents.

Their findings showed an interesting disconnect between coding activity and completed software.

After the adoption of AI coding agents:

• Lines of code grew by 30%. Developers were writing much more code.
• 20% increase in code commits Engineering teams said they’ve made more updates to their projects.
• Pull requests up 23% More code was coming up for review.
• 49% increase in code review time. The process of evaluating and approving code was much longer.

Perhaps the most important finding was the fate of completed work.

Perhaps the most important finding was the fate of completed work.

Despite the increase in coding activity, the researchers did not see a statistically significant increase in completed software output at the company level.

The study also found that the proportion of pull requests that required changes almost doubled and review comments increased by 35 per cent.

These findings were reported in an Ars Technica article on AI coding productivity published in October 2026.

The problem is not AI’s poor performance on coding. Not at all. The technology certainly makes it easier for developers to do some things faster.

The trick is when the code is written.

Someone still has to review the output, find mistakes, test functionality and decide if the software works the way it is supposed to.

These limitations were observed by the researchers in the firms and time period they studied. Their results don’t set the ceiling for what all future artificial intelligence systems will be able to do.

But they do make an important point about technical work: Getting an answer isn’t the same as knowing if the answer is right.

So we get to a question a lot of folks are asking.

Will Software Engineers and Programmers Become Obsolete Due to Artificial Intelligence?

So it’s no surprise that searches around will AI replace software engineers and will AI replace programmers are so popular.

AI coding tools are rapidly improving. They can suggest functions, identify bugs, explain programming concepts, and even write large chunks of applications.

As these tools get better, some programming tasks will almost certainly become less manual.

But software engineering is much more than just writing instructions.

Let’s say a company builds an application to store customer data. An AI assistant can write the code for the login system in a matter of minutes.

That doesn’t necessarily tell the company whether the system is secure, how it should handle unexpected user behavior, or whether it integrates well with the rest of the application.

These are questions of technical understanding, testing and informed judgment.

The same can happen with an AI-generated program that seems to work but has a bug that only shows up under certain conditions.

Experienced developer needs to realize the problem and realize the implications and decide what needs to be different.

Of course, AI may also increasingly assist with these tasks. The balance between human and machine labor will continue to change.

But the Harvard research is a useful reminder that software development is a process with several interconnected stages. Just because one stage is faster, doesn’t mean the work needed somewhere else disappears.

The wider employment data is also worth looking at.

According to the U.S. Employment of software developers, quality assurance analysts, and testers is projected to grow 10% from 2025 to 2035, faster than the average for all occupations.

Those projections don’t guaranty that every programming job will stay the same. However, they do contend that the need for software-related expertise cannot be simplified to the number of lines of code an AI system can generate.

Perhaps a better question for families and educators is: What should students learn to participate meaningfully in a world where AI is part of the development process?

Why Human Problem Solving Still Counts When AI is able to Code

Picture a student working on a basic robotics project.

The purpose is to have a robot programd to go round an obstacle and reach a specified point.

An AI tool could tell you the code you need to control the robot’s movement.

But the first try doesn’t work.

The robot turns too soon, misinterprets a sensor, or goes too far.

Now the student is faced with a problem that cannot be solved by adding more code to the screen.

They need to look at what happened, figure out what might have caused it, try a different approach and see how it plays out.

That process demands several skills that transcend programming itself.

Critical thinking requires students to challenge an answer rather than taking it at face value.

Problem solving encourages them to think of alternatives if their first try is unsuccessful.

In engineering design, they learn how to develop solutions through designing, testing, and improving from their observations.

Communication and teamwork: They can share their ideas, listen to feedback and work with others.

The skills don’t lose their relevance just because artificial intelligence is able to generate instructions.

Actually, access to powerful tools makes it especially useful to understand the results of them.

This is why coding education shouldn’t be about memorizing syntax or learning to type commands.

Students need opportunities to make sense of what they’re building, why it works, and what to do when it doesn’t.

iCode covered this concept in an earlier article, Why Learn to Code When AI is able to Write Code?

The educational argument still stands but the new research adds an important real world dimension: even professional engineering teams have to cope with the gap between generating code and delivering a working solution.

What does this mean for STEM’s future?

For years, learning to code was often sold as a way to prepare for a potential career in software development.

This is one possible outcome, but the value of STEM education is far more.

Logical reasoning, structured problem-solving and the ability to break down a complex task are also required by a child learning to program.

Robotics brings those lessons into the physical world, where designs have to deal with motion, sensors, materials, and real-world constraints.

Game development involves creativity, interaction design, testing, iteration.

AI education brings in another important dimension: learning how to use smart tools and challenge their accuracy, constraints and proper use.

These are valuable experiences for students, even if they do not become professional programmers.

That doesn’t mean the goal is to get each child ready for one technology job. It’s about giving them experience of understanding and working with increasingly complex systems.

It also means education providers need to rethink how they conceptualize successful learning.

Or if a student asks AI to build a website, the result can be impressive. But what has the learner actually learnt?

Can they explain how the site works? , see problems with its design? Can they change how it works or make it better for someone who is using it?

Those questions tell us much more about understanding than about whether the original code was written by hand.

For an education business, the difference matters.

If your curriculum is designed around teaching a tool, you risk the tool changing and you becoming obsolete.

A curriculum that builds technical foundations and introduces new technologies has a different goal: helping students develop capabilities they can apply across changing platforms and environments.

That’s the kind of educational approach to consider when contemplating the long-term relevance of a STEM education business.

How iCode Prepares Students for an AI World

Learning with iCode is more than simply following coding instructions.

Students learn technology through a progressive curriculum that combines programming, robotics, engineering, game development, design and AI-related concepts.

Emphasis is on giving students chances to build projects, overcome problems, and hone both their technical and people skills.

For example, iCode’s White Belt introduces robotics and programming using VEX IQ. Students learn about robotic systems, engineering design, electricity and 3D modeling as they work their way to more complex robotic projects.

A student is not just learning how to enter commands. They are learning how the decisions they make in programming affect a physical machine, and how to change a design when reality does not match what was originally planned.”

In Orange Belt, students get to learn web design with HTML, CSS, and JavaScript. They learn about graphic design, photography, and user experience principles.

Building a website is a process of making choices about how information should be arranged, how people will use the finished product, and if the design does what it is supposed to do.

Meanwhile, Green Belt introduces students to Python, drones, Arduino, data concepts and machine learning.

Students in these programs have the opportunity to interact with different technologies and continue to build their understanding of how systems work.

In addition, iCode intentionally folds in AI principles into its proprietary curriculum such as responsible use of AI, prompting, evaluation and reflection.

Instead of positioning AI as a substitute for foundational knowledge, this approach frames it as a tool that students can learn to understand and use thoughtfully.

Another key feature is the instructor-led learning environment of iCode.

Students work on projects, ask questions, collaborate with classmates, and troubleshoot challenges with guidance.

That human interaction provides instructors with the opportunity to foster greater understanding rather than simple acceptance of a completed output.

That broader approach to technology education is reflected in iCode’s Belt Program and its STEM.org-accredited curriculum.

The company’s curriculum also includes soft skills such as communication, teamwork, critical thinking and project management.

This is a distinction that matters to a prospective franchisee. The educational offer is not confined to a single programming language or a single category of technology.

It mixes a variety of technical topics with skills that will continue to be useful as the technology itself evolves.

Why Should You Invest in an iCode STEM Education Franchise?

The Harvard findings help explain why technical education is relevant, but they don’t tell us whether a particular franchise is a good investment.

And that’s another question—a question that requires some thought.

If you are an investor exploring education franchise opportunities, you need to consider the curriculum, business model, customer needs, operating requirements and adaptability to market changes.

There are a number of features of iCode worth considering.

A curriculum that goes beyond traditional coding classes

The education model of iCode includes a wide range of technical disciplines such as coding, robotics, engineering, game development, and AI-related learning.

It’s progressive Belt Program also provides a structured learning path instead of just relying on isolated workshops.

“That provides families with opportunities to have students continue to develop skills at different stages of their education.”

It offers franchise owners a host of educational offerings to discuss with potential customers.

Different learning needs have a lot of programs

An iCode campus is not built around one class.

The company’s franchise income streams include ongoing Belt Programs, summer and holiday camps, Tech Clubs, events and other educational experiences.

Some programs provide ongoing learning experiences throughout the school year. Others expose students to new topics during school breaks or in partnership with community organizations.

These formats provide multiple options for families to engage with iCode.

They can also create multiple sources of income for franchisees but results will depend on enrollment, local demand, costs and execution.

A curriculum that evolves with technology

Technology education cannot be stagnant.

New programming environments come online, AI tools get better, and student interests change.

iCode describes its curriculum as “always evolving,” with input from technical specialists, teachers, franchisees, families and students.

That’s especially the case when dealing with an education business focused on technology.

No curriculum can foresee all developments in AI. But for future owners, a practical consideration is that they will have an established process for reviewing and updating educational content.

A franchise structure that’s recognized to assess

Prospective owners are investing in more than curriculum. They are trialing the systems required to operate an education business.

More on the iCode franchise business model. The company’s educational offerings and campus operations are built on a franchise business model.

Those details should be considered alongside the initial investment requirements, ongoing costs, market fit and the franchise disclosure document.

It’s one thing to be a nice educational concept, and another thing to be a good personal investment decision.

That AI is getting more capable doesn’t mean more students will flock to STEM fields. Also, having lots of programs does not guaranty profit.

What the research does offer, however, is a strong argument to look beyond the assumption that technical education is rendered unnecessary by AI-generated code.

If you’re an investor who believes in the value of learning through doing, it’s worth looking at how an education business teaches its students to work with technology, not just follow it.

The Future is Not Just Who Can Write Faster Code

Artificial intelligence is changing software development, and those changes will keep coming.

Some things will be easier. Some can be almost completely automated. There may be new ways of working, and new responsibilities.

But it’s still important to understand a problem, evaluate a solution, and improve something that’s not working.

The Harvard research provides a good example. There was more code written but it still needed some love in the form of review and other stages of software development to get it done.

It teaches students the importance of learning for learning’s sake.

For entrepreneurs it poses a different question: what kind of education business is worth building when technology is always changing?

iCode’s methodology integrates a practical technical education, a dynamic curriculum and a franchise model based on several educational programs.

If you’re looking at a STEM education franchise, the next step is to find out how that model works in practice.

Find out more about the iCode Franchise Opportunity and contact the franchise development team to learn about ownership requirements, opportunities and the process for opening an iCode campus.  

References

1. Chen, F., and Stratton, J. (2026). Artificial Intelligence in the Firm: Bottlenecks in Software Production (working paper).

2. Orland, K. (2026). AI coding agents generate more code, but not more software. Ars Technica.

3. U.S. Bureau of Labor Statistics. Software Developers, Quality Assurance Analysts, and Testers. Occupational Outlook Handbook.

Empowering Young Minds, Inspiring Future Innovators.

iCode Franchise