AI is getting better every day.
As a software engineer, this is a topic that is often top of mind, which prompted me to share the plan I've put into action to keep an edge over AI.
I believe it is a useful trait to be paranoid, within reason.
Leveraging new tools and technologies to remain relevant is arguably the nature of our job.
Disclaimer that this is not a plan to save the world (I wish I had all the answers!) but rather my personal set of actions I believe will increase my odds of maintaining an edge over AI.
At least for a little while longer...
Here is a visualization of my plan. We will cover each part in detail.
Foundations and fundamentals
Before we dive into my actual plan, let's start with a short story.
The story of how I became a software engineer.
If someone had told me 10 years ago that I would be working as a software engineer today, I would not have believed them.
I was at university studying business and finance.
After graduating, I started interviewing for jobs in finance. In parallel, I was experimenting with drones, which were a hobby of mine.
I quickly realized that there was an opportunity to package drone services to realtors to promote their luxury properties via aerial imagery and video.
I learned a lot in the process from cold calling, marketing, financial planning and account management.
But above all, I learned that I wanted to build software. We had clients asking us for tools to edit our media assets, store, stream and serve them to their clients. This got me digging into the internals of software and opened my eyes to numerous possibilities.
Shortly after, I signed up to a coding bootcamp to learn web development fundamentals.
I struggled during 9 intensive weeks, but it was transformative in the best way possible.
My brain was challenged to think in new ways: data structures and patterns, software design and components, principles and frameworks.
I haven't stopped coding since.
6 years of building software, from startups to larger companies.
I recently started a new job as a software engineer and noticed how my ways of working have drastically changed since my first lines of professional code.
I am not saying things are better or worse.
Just like my irregular path, each step brings along its set of challenges but also rewards.
Change is the only constant, and with it come boundless opportunities.
I wanted to open with my story to encourage you to embrace change and also emphasize the importance of investing in fundamentals.
This is foundational to what we will cover in the next parts.
Unlock potential with the right mindset
An optimistic mindset is key to unlocking potential and working alongside AI in a positive way.
In some cases, there is too much pessimism surrounding AI.
I don't believe this is healthy.
I am not saying we should blindly trust AI, but we should accept it as new tooling that can be leveraged and start getting used to it as a new norm.
I acknowledge that the job market is tough, especially for junior developers.
Stay curious, passionate, excited, open-minded.
Within my network, I have witnessed juniors with no professional experience be very creative and resourceful, and land great professional opportunities.
I highly recommend reading the post Hunger Beats Tenure written by Felix Rieseberg, a leading engineer at Anthropic. The post conveys a refreshingly optimistic take for junior developers with practical actions to take to stand out in the current job market such as leaning into your fresh perspective and aiming for radical ideas with big potential for change.
Pick a development plan and stick to it
Think hard about your personal goals and build a development plan to reach them.
In this attention economy, people tend to swipe from one thing to the next, chasing one hot topic after another.
There is no harm in staying on top of industry trends. However don't make this your core strategy.
This age of AI rewards both generalists and specialists. The former can create associations across tools nobody has considered and the latter can unlock deep intuitions.
I personally like to follow a T-shaped development approach to get the best of both worlds.
The horizontal bar of the T represents my efforts to stay on top of the latest trends and experiment, whilst the vertical stem of the T is me deepening my knowledge within fintech engineering. This is where I spend most of my learning and development time (70-80%).
Follow your plan rigorously. It might feel slow at the start but once you pick up momentum you will feel the benefits when the dots start to connect.
Deepen domain expertise
My focus is on deepening my knowledge within one specific domain to gain invaluable experience and industry knowledge that will make you stand out from the crowd.
If you can combine deep knowledge at a business level and translate that knowledge into technical solutions, you will be perceived as unique and valuable.
In my case, I am developing an edge as a fintech engineer.
I have been working in the space for almost 5 years now and understand how payment flows and systems work under the hood.
I stay on top of the latest trends through various resources such as books borrowed from my local library, podcasts and newsletters and through community meetups.
If you are interested in fintech, I personally recommend the following books as a start:
- The Anatomy of the Swipe
- The Pay Off
Broaden your awareness of tools
Become AI-native, not AI-dependent.
Know when to leverage AI appropriately by making use of available tooling.
AI is powerful for information retrieval, exploration and synthesis.
In this sense, context management becomes critical.
Here are some practices that have helped me work more effectively with AI:
- invest more time upfront gathering requirements and discussing requirements with stakeholders, and collecting these findings in a living document your agent can ingest
- embrace spec-driven development to structure your AI output
- invest your time writing detailed Architecture Decision Records (ADRs) to capture the why not just the what
- connect your tools and sources of information through MCP servers and connectors to plug your agent across all critical tools (Jira, GitHub, Slack, etc.)
- explore broadly with AI before going deep with your judgment
- use AI to be more proactive with stakeholders by sharing prototypes rather than just documents to convey ideas better
- leverage AI to be more helpful to others. For example, last week I used Claude to generate a Go guide for my team's onboarding
Minimize AI slop
Leveraging AI is good, but using it deliberately is even better.
Just because AI makes it easier to produce more doesn't mean you should.
When joining a new organization, don't jump in and make changes immediately.
Learn the engineering ways of working first before making any changes.
With AI-driven development, if you don't understand the rules and define scope, the blast radius can be high.
Some practices I adopt at work:
- scope PRs to smaller changes to make reviews easier and limit potential issues
- think before prompting by writing down your thoughts, assumptions and approach
- introduce human checkpoints (architectural decisions, big change reviews)
- when prompting, define your target audience to increase relevancy
Before sharing AI-generated work, I ask myself:
- Have I actually reviewed it? A simple rule should be: don't share a document with others that you wouldn't read yourself
- Am I contributing original thinking? Personal insights, experiences, perspectives.
- Am I saving people time or costing them time? A lengthy AI-generated document can take minutes to generate but take hours for colleagues to interpret.
- Am I confident signing my name on the AI work? Accountability is key.
- Am I using an AI disclaimer? Explicitly mention when an artifact is AI-generated so consumers can apply extra scrutiny when reviewing.
An informal way of measuring the above is to ask yourself: are you someone people enjoy working with who provides value?
You can go a step further and drive value at an organizational level.
From experience, AI is not good at driving change at that level effectively and with positive feedback from engineers.
AI tends to overengineer. I always appreciate an engineering culture that solves complex problems with simple solutions and effective tooling.
Personality beats information
With information becoming increasingly accessible, personality, experiences and unique perspectives become more valuable.
I invest more time in developing my soft skills that make me a better collaborator:
- communication
- listening
- gathering requirements
People tend to value an experience more than just the solution.
This post on X conveys this idea nicely:
Don't let AI take away your opinions, the fun and joy of building.
I never stop building and experimenting.
I've noticed that more companies seem to recognize that in the age of AI, a strong recruiting signal is whether the candidate can entertain a passionate discussion about building software systems.
Passion and excitement are behind the best software. This contributes to the experience provided.
Instead of beating AI at what it does best, I try to beat it where it performs worse: passionate and creative thinking derived from real-world experiences and unique perspectives.
I spend time building my network. To be part of a community that challenges and sharpens my perspectives.
AI can't replace the raw sharing of human experiences.
I actually attended a meetup the other week with the theme 'What can't AI do?'
Amongst fellow engineers, we spent hours debating and sharing our unique perspectives.
I aim to elevate and share my unique voice.
To build a personal identity and contribute to the conversation.
Some ways this can be done:
- write on DEV by sharing your experiments, thoughts and discoveries
- engage on LinkedIn and connect with others
- maintain a portfolio of projects on GitHub that reflect who you are and what you are passionate about
It is not only the what you are building that is important; it is the why.
My question to you is: are you also working on your own plan? Which skills are you currently investing in?
If so, I would love to know more! Please share any useful tips with the community on how you are maintaining an edge over AI.
If not, it is not too late to start working on one! Let me know in the comments if you require any advice, I would be happy to help you get started.
Top comments (24)
Thanks for this! I have been struggling to think of a plan that's best for me and this is very helpful guidance. It's so easy to get lost with all there is out there and overwhelmed, too! But your staged plan makes it less so.
Thanks! Glad to hear.
I wanted to write out and share my plan in the most practical way possible.
Hi,
immediately caught by the title, since i have been working on a similar post about ai and how i try to handle it.
It is currently on hold, i am very busy and i personaly need some more time to process it.
I am currently trying to handle all this AI hype and try to find my way. Tech changes too fast to process it in the way we used to do.
I am not very optimistic. Try not to be pessimistic.
happy to read your post and know about you
Hi, happy to connect with you.
I look forward to reading your post and learning from your plan, when you release it.
That's a smart, well-thought-out plan. Thanks for sharing!
Thanks Sylwia! Your post from the other week actually inspired me to share my plan here.
Great framework, especially around understanding underlying systems instead of just consuming higher-level APIs.
One concrete way I've stayed sharp with AI is running and tuning models locally (via Ollama/llama.cpp) rather than treating AI purely as a cloud prompt box. When you actually see how KV cache scales, why context window growth eats VRAM non-linearly, and how GPU offloading splits layers across memory buses, you understand what AI tools can and cannot reliably do in production.
For anyone experimenting on consumer hardware, keeping context windows lean (e.g.,
num_ctx: 4096) and monitoring layer offloading gives you surprisingly capable performance on zero monthly subscription burn.How much of your day-to-day workflow are you running through local tooling vs cloud LLMs?
That is a great point I didn't mention in this post but I have been more deliberate about my choice of including open models I can run locally vs cloud LLMs. This has pros and cons over privacy, performance, transparency as you mention.
Your T-shaped split is the part I want to copy. Spending 70 to 80 percent of learning time deepening one domain while still experimenting broadly is a much clearer rule than just staying curious, and fintech is a strong vertical for it. The ADR habit stood out too, since writing down the why is exactly what keeps AI output reviewable instead of just fast. And the question about whether you would sign your name on the AI work is a great checkpoint. I am going to borrow that one before sharing anything generated.
Nice to see AI comment and acknowledge my plan to beat AI ;)
The obsession with wrapping remote APIs into multi-layered agent abstractions is creating massive hidden latency and failure cascades in production.
If you profile real-world workloads, bare-metal local inference (via direct C++ / llama.cpp or vLLM bindings) consistently beats fragile cloud wrappers on both predictability and cost. Tooling should eliminate abstraction penalties, not compound them.
This does not seem relevant to the topic I posted about :)
The transition from business/finance to software is actually the exact skillset that AI commoditizes least.
Writing syntax, scaffolding boilerplate, and prompting models is trivial now. What models fail at is evaluating whether a piece of software is actually solving a commercial problem that someone will pay for versus 'selling furniture to carpenters'.
Engineers who understand unit economics, conversion bottlenecks, and domain friction of real businesses will always run circles around pure syntax typists.
Speaking business and product language and knowing how this translates well into technical requirements definitely can be an edge today.
I think I came across this article at the right time, Julien. I was actually feeling a little overwhelmed lately, thinking about how things are changing and what I should be focusing on. Reading through your approach gave me some guidance and helped me think about things a little differently. I think I needed to read this. Thank you for sharing 😀
Nice to hear from you Hemapriya! It does feel like a rollercoaster at the moment. I am glad you found some guidance in my post. If there is anything you need more clarity and advice on, feel free to reach out to me :)
Strongly agree with your emphasis on fundamentals over prompt engineering.
One practical shift that gave me massive clarity was moving away from cloud AI black-boxes and learning to run/quantize models locally (llama.cpp, Ollama, GGUF).
When you understand where memory bandwidth actually chokes (KV-cache footprint vs layer weights), prompt evaluation vs token generation, and context truncation, you suddenly see through 90% of the AI hype cycle. You realize these aren't magical thinking engines—they are memory-bandwidth-bound matrix multiplication graphs.
That foundation completely changes how you build software on top of them, whether you're designing local agents or reducing cloud API bills.
Good point. I am trying to be more deliberate on my choice of open models vs closed.
reassuring, thanks coach.
You are welcome. Do you also follow a plan?
I follow email subscriptions to see what's happening in the ai space . but I try to avoid ai psychosis the best I can thru a good ole book.lol
Agree, it’s hard to beat a good book!
Any recent recommendations?