How is project execution changing in 2026

I think writing projects or executing them in the past involved three major steps, so usually the project goal came from the business metric that leadership wanted to improve for example it could be increasing revenue, drive up user traffic or just show more ads to boost sales.

Once you had the idea and it got approved you had the following things happen most of the times in the order below

• High level design document outlining the features

• Once the design is ready get the PMs and engineers aligned on the document and the method of implementation and the teams involved in order to get it done

• Set up a weekly cadence and set goals for each week

• This is usually when the engineering started and the points we are going to focus on largely affect this area the steps above are still mostly the same regardless of AI .

  • Engineers read the existing code and start to modify it to incorporate existing features

  • Spend time putting up diffs and getting engineers to review and approve.

  • Now senior engineers played a huge role here because they had the context of what architectural design decisions were taken, who modified what and where the mines were buried and how you could avoid tripping them up. Overall they had a great understanding of the codebase in general

  • Iterate for the above steps for a month and you had a working feature set, with monitoring alerting and the usual SRE stuff that comes along with writing a new feature.

  • You then send it to QA and after a few rounds of feedback you could launch, easy peasy lemon squeezy.

• But AI flipped the script on its head

  • The step where i spoke about how an engineer reads the existing codebase and puts up diffs, yeah that part is completely taken over by Large Language Models, they now read the codebase, suggest changes and put up change requests

  • The engineers are merely in the loop, all they have to do is go through the changes see if it works, scroll social media and hit send for that 10th PR of the day

  • The reviewing is equally bad, remember how i talked about how senior engineers had an understanding of the codebase because of the huge muscle memory of the codebase, yeah that is gone too, because with the increase in pull request volume senior engineers simply dont have the time to take a look and review diffs turning that metric on its head too.

  • If it works it ships: This is the attitude thats prevalent now because lets face it we all have the pressure of delivering on tight deadlines so now one is actually reading the code line for line if LLMs can do just as good a good as a normal engineer can.

So what has that lead to ?

If you look closely enough the understanding that the engineers had of the codebase in the past does not really exist anymore because it has been consumed by lending that understanding to the LLMs or rather outsourcing the thinking to the models to do. Which leaves projects in a position where no one really understands whats happening why ? because remember when you hit approve on that pull request ? yeah that one unfortunately there was a piece of code that caused it to break.

No worries though. The LLM will fix it for you and you have 5 more minutes to scroll on social media.