How AI Will Change Software Engineering in 2025

Observations on What Could be Coming for Product and Engineering

Is AI only a semi-competent junior programmer that likes to hallucinate or is it quickly taking over every role in Product & Engineering? 😱

Despite the cool demo’s from companies like OpenAI, there is still no question that in most organization AI is NOT taking over and replacing roles at scale this year.

There are several reasons for this:

• AI is not very good at writing code, writing requirements, etc

• People don’t know how to use AI properly yet (e.g. prompting, RAG, etc)

• Companies are ALWAYS slow in technology adoption

• Regulatory, privacy and compliance concerns are a big thing in a lot of industries

• AI thrives on data and a lot of companies have poor quality data

OK, so AI isn’t taking over Product & Engineering roles TODAY but what about the future?

Well, technical innovation in AI will of course continue along a similar curve in 2025 as it has in the last couple of years. The money going into it is too big. The technical talent working on it is too good. And the whole world is so focused on it (check out DeepSeek).

Of course that doesn’t mean organizations will actually adopt AI with speed — the adoption part will take much longer. Companies are notoriously slow at adopting tech of any kind and so far AI adoption has proven to be slow in the enterprise (despite its overall popularity.)

So there’s AI innovation and then there’s AI adoption — two different things.

Let’s use both of those lenses and look at the likely impact of AI in each big functional area of Product & Engineering in 2025.

Product Management Impacts

Product Management is a broad set of responsibilities but at 90% of organizations these days the job boils down to these 5 core things:

• Writing requirements and explaining them to engineering

• Talking to customers, sales/marketing, and support teams

• Prioritizing the backlog, cleaning up tickets, and deciding on the next sprint

• Presenting roadmaps, negotiating timelines, figuring out release dates, etc

• Tracking metrics & KPIs

Writing Requirements

AI can seriously enhance the ability to write requirements because LLMs are well suited for the task, so PMs will gradually do less & less of it.

However, right now AI is not very good at it. It’s bad at understanding the context of the product, the customer needs and so forth — I’d give it a C-.

And of course, explaining requirements to engineering is a long way off. You need a ton of context to do it well and human PM’s are best suited for it at the moment.

But I could easily see an AI agent trained on a companies data-set over time be able to do the “explaining” work (even verbally) sometime in 2026.

Companies of course would love to adopt this kind of AI because it means scaling the impact of their PM team, but like I said the tech isn’t there just yet to write great requirements.

Talking to Customers

AI won’t directly talk to customers in 2025 either. But it is already playing a major role in customer interaction, and this will only grow.

Tools like sentiment analysis software ARE for sure helping PMs gather insights from customer feedback at scale today.

Also, instead of scheduling multiple customer interviews or sifting through 100’s of support tickets, PMs are relying on AI to summarize themes, find pain points & suggest feature ideas.

Salesforce, HubSpot and other enterprise tools are also building AI features very quickly which will make it easier for PMs to have the right data when talking to Sales & Marketing.

In 2025 all the actual conversations between PM and other stakeholders still have to happen human to human, but AI is getting better at supporting this work.

Prioritizing the Backlog

This will be difficult for AI to do for a while because the work requires a lot of human instinct, and contextual understanding that AI lacks.

Backlog prioritization has always been a balancing act of business needs, customer demands, and technical constraints — AI is simply below average at that.

But, AI will make this process faster and smarter for humans in 2025.

Tools are already emerging that analyze metrics like feature usage, defect rates, and revenue impact to suggest backlog priorities & innovation here will continue.

But human judgment will still be critical, especially when weighing “soft factors” like team morale, stakeholder politics or just an annoyed customer that needs a fast fix.

The adoption of this kind of AI will happen through existing tools like JIRA. And smart PMs are already experimenting with using these & even uploading their backlog to OpenAI, etc.

Building Roadmaps

AI can generate nice looking roadmaps based on inputs like priorities, resource availability, and deadlines — but there’s not much thinking going on behind the scenes so they are just OK.

Roadmapping tools like Aha and other products are adding AI into their offerings faster and faster every day so the changes are coming.

But AI is not negotiating timelines with other stakeholders anytime soon. You’re still going to need humans to do that in 2025.

What’s coming next from AI is exciting though: analyzing past roadmaps, identifying dependencies, predicting risks — then using that to recommend release dates, scope, etc.

From an adoption standpoint tools like Aha & JIRA are building AI-based roadmapping features and PMs will be using these tools quite a bit in 2025.

Tracking Metrics

AI is pretty good with helping track numbers. In fact it’s an amazingly helpful tool for setting up KPI dashboards and piping data into them.

So PMs will have to do less of this by hand over time.

PMs will eventually rely on AI to aggregate, analyze, and generate insights from KPIs in real time — lots of enterprise products are building these features.

This means fewer spreadsheets to manage! 😃

Adoption here will be very strong in most companies. In fact, it’s already happening in PM teams everywhere. A big chunk of PM work will get easier because of this.

Engineering Impacts

OK, let’s talk about Engineering. A lot of people are worried about AI’s impact in Engineering in 2025. Here are the areas we’ll look at:

• Writing code

• Testing / QA

• Running CloudOps

• Designing systems (architecting)

• Making estimates & talking to Product

Writing Code

AI is a junior developer at best right now.

But it will keep making inroads into the repetitive, boiler-plate programming that goes on in a lot of companies. Tools like Copilot will keep getting better and really help here.

I think for the more sophisticated and complex code AI is not going to help a ton in 2025 unless we see a massive breakthrough in the middle of the year sometime.

Companies are salivating about adopting AI coders but the technology has performed fairly poorly in real world situations — like I said, junior developer at best.

Right now AI helps programmers mainly by assisting them in researching technical problems and by providing simpler, boiler-plate code — none of which is game-changing.

Testing / QA

I’m sorry but testing and quality assurance are on the brink of a **savage reckoning** as AI transforms this area more radically than almost any other in engineering.

AI is perfectly suited for QA because it thrives on routine, repetitive tasks like generating test cases, identifying bugs, and running regression tests.

These tasks, which historically consumed countless hours of manual effort, are now being automated with remarkable speed by AI-based testing companies/products.

Tools were already emerging that could analyze code for vulnerabilities, simulate edge cases, and even predict areas of the codebase most likely to fail—even before ChatGPT came out.

The result?

A complete upending of the QA landscape. In fact, I think manual testing will be almost completely extinct in 2 or 3 years. Even automated testing will be handed over to AI very quickly.

Companies are adopting this kind of AI so rapidly in 2025 because it’s so effective for replacing certain human workflows.

Running Cloud Operations

CloudOps presents a unique challenge for AI because the work involves a mix of automation-friendly tasks and nuanced, context-heavy investigative work.

The big cloud providers like AWS will continue to release AI-based features that help CloudOps Teams do their work more efficiently.

But it’s a far cry from replacing a CloudOps Engineer anytime soon.

For example, AI isn’t going to be able to investigate some really unique problem on a random server the way a human can right now.

I think CloudOps Engineers better be using AI pretty extensively to keep up with the tech or risk getting left behind, but the tech can’t do what they do yet.

However, Cloud is a major expense area for companies, so you better believe they will be looking to adopt more and more AI-based solutions here.

Still, I think the biggest threat to CloudOps Teams is not AI itself but a new crop of cloud providers recently popping up that use AI to offer better services than AWS & Azure can.

These “Next Gen” AI-native cloud providers might make it so easy to run your applications in the cloud that a 10-person CloudOps team may shrink down to just 1 or 2 people.

So AI’s impact on CloudOps is much less straightforward.

Designing Systems (Architecture)

Big fail here for AI so far. Not that it won’t make progress in 2025. But getting system architecture right is the job of a senior engineer and that’s not AI right now.

Of course, AI will give architects superpowers over time.

For example, designing scalable, efficient systems often requires identifying patterns from a sea of prior knowledge, assessing trade-offs, and predicting outcomes.

AI will get pretty good at that.

But mass layoffs of Architects isn’t happening because of AI in 2025 — not unless there is a huge breakthrough at OpenAI, Claude, DeepSeek or Gemini on this front.

AI right now can be a friendly & reasonably smart colleague to bounce ideas off of — but that’s about it.

And from an adoption standpoint, companies aren’t running out to purchase AI for architecture purposes. Architects are expensive but they aren’t a big % of the payroll, typically.

Making Estimates

AI will (eventually) be good at the act of making engineering estimates. That’s because AI can do pattern matching and look at historical data very well.

Traditionally, estimating timelines, resources, and effort has relied heavily on human intuition, historical knowledge, and gut instinct.

But this approach is notoriously prone to underestimation, missed complexities, and costly overruns — we’ve all seen it!

But even with AI doing this work better than people, you’ll still need humans to spend the hours to debate timelines and tradeoffs — that parts not going away.

Adoption for estimation types of AI tools will be fairly high for companies via enterprise products like JIRA because it can mean so much potential recovered budget. 💰

Final Thoughts

Let’s recap!

In Product Management in 2025:

• AI WON’T write good requirements yet & won’t explain them to engineering

• AI WON’T replace PMs in talking to customers, but will help them with data

• AI WILL start doing more backlog management

• AI WILL start do more roadmapping but won’t negotiate timelines yet

• AI WILL do a lot of the KPI tracking that PMs do by hand right now

In Engineering in 2025:

• AI WON’T write a bunch of great code unless there’s a big breakthrough

• AI WON’T replace CloudOps engineers, but “Next Gen” AWS competitors might

• AI WON’T replace Architects

• AI WILL completely torpedo the testing and QA field (sorry 😥)

• AI WILL do a lot of engineering estimations

So, AI will change & even enhance Product & Engineering in 2025 across the board, but some areas will be hit harder than others in terms of total transformation.

In Product AI is not good enough to write requirements well or talk to customers, but it’ll start doing backlog management, strongly support roadmapping, and do a lot of KPI tracking.

In Engineering AI is not good enough to write great code, do CloudOps work or replace architects, but it will probably kill QA and will be doing a lot of engineering estimation work.

So PMs and developers aren’t disappearing just yet.

But, there are some scenarios where AI will definitely hurt certain people in P&E:

• If you’re mediocre and want to stay that way you’re in trouble

• If you don’t learn the AI toolsets you’re in trouble

• If you reject AI as a concept or view it as your enemy you’re in trouble

So, one of its biggest impacts of AI will be to exit the lower D & F talent out of the industry. Those individuals will no longer be able to succeed in P&E.

This will then probably give way to a new “partnership” model between humans and AI when it comes to building software.

Humans will have to learn to:

• Think & reason together with AI

• Communicate & collaborate with AI

We didn’t have to do that with technologies like Cloud or Blockchain, of course.

But AI is different — we are inventing a new, interactive partner to work with us to create products.

And all Product Managers & Engineers will have to contend with that from now on…whether they like it or not!

Reach out for help if you need it: bobby@technocratic.io.

And in the meantime, keep the shark swimming! 🦈