Table of Contents
I’ve been talking with a few startup friends, and they are telling me the same story: “We are not hiring anymore, only some specific roles. Thanks to AI, we are now 2-3x more productive.“
They are thinking: “Instead of hiring a new engineer who could risk my current speed, I’d rather invest in AI to make my team go faster.”
After all, why spend $100k a year on a Junior Engineer when I can pay $1k a year to make my Senior Engineer twice as productive?
The problem? Everyone is doing the same.
The whole industry is now 2-3x faster; so they are in fact not moving fast, they are moving at the same velocity as everyone else.
Are you really moving fast?
We are in the midst of a technological revolution – THE technological revolution.
I understand everyone is blind to how incredible this is: finally, there is a platform that can write code, not the flop that GeneXus has been for more than 30 years (which now is powered by GenAI).
The common feeling is a mix of “this is more than we hoped for” with “let’s be cautious.“
Who would be crazy enough to risk adding a non-AI-proficient engineer or to try to stretch AI a bit more for a meaningless speed gain?
Let’s wait and see…
If you started using LLMs to generate code back in 2023, the model and tools you are using today are very different.
There is a clear pattern: a new model comes up, the previous model becomes cheaper, the development tools company integrates the model natively, and your productivity keeps going up.
Now, the market is talking about development Agents (like Devin) who could potentially replace an entire development team.
So there is no point in risking this heaven; let’s just wait.
Buy vs. Build
This is not the old Buy vs. Build dilemma — startups don’t even consider setting up their own servers these days (unless there is a good reason); they will use any Cloud provider. The reason? Cost.
Why is it different in the AI era? Because, for the first time, you can do both: you can buy models, you can build your own model. You can buy a Copilot subscription, or you can build your own Copilot. It doesn’t have to be expensive. It is actually cheap.
Of course, many startups crash and burn because they’ve built their entire model around a single AI feature that, the next day, a big company like OpenAI will launch and completely wipe them out.
But that’s not your case, I hope. Your moat is not AI itself. I hope.
AI is so easy to integrate (as simple as one decent prompt wrapped up in a script) that the cost of automating or accelerating a task will be easily recouped.
A few ideas to increase your -average- speed
Here are a few ideas that, even if you spend a sprint or two, will drastically increase the speed of your team:
- Data quality and integrity validator: If you deal with tenant data, you want that data to be clean. A simple event-triggered or batch process with a set of cleaning rules to spot potential inconsistencies could be a game-changer. Managing multiple databases? Validating data consistency across resources is even more powerful.
- API or Library breaking changes: We all experienced annoying API breaking changes or non-backward-compatible library updates. Usually, third-party providers don’t communicate the changes well, and they won’t analyze your codebase to tell you exactly what to change. The solution? Inject the changelog as a context in a prompt that analyzes your source code.
- House-rules linter: One thing that GitHub Copilot won’t do is follow your own code guidelines and architectural rules. You can make those part of the prompts, but it is easier if you build a simple script that will compare the code with your proprietary best practices.
I won’t get into QA automation, Security audits, or DevOps monitoring — those problems are generic enough that Microsoft or JetBrains is most likely already working on AI-powered solutions.
If the problem is specific to your startup, there is an opportunity.
Are you afraid of moving faster?
A wise man told me: if you want to drive fast, you need good brakes.
Do you have good brakes? AI is no different from any new technology: you need ramp-up time and a learning curve to actually see ROI.
The good news is that it is flexible enough for you to allocate just one day and see a return the next day if you know what you are doing.
The technology is so versatile that a good prompt for a repetitive task wrapped in a Python script triggered by a scheduled task can increase your development speed.
Ask yourself this: Do I want to ride the wave? Do I want to do the same as everyone else? Or should I take this opportunity to move faster than average?
I assure you, other startups are going the extra mile with AI to increase productivity – they did not stop hiring: they keep adding more engineers to help them implement AI, both operationally and at the product level.
How do I know?
Because some of them are my clients at InTheValley.
Are you brave enough to ride this tech wave at full speed? Then talk to us!
- Stop Consuming AI and Start Riding AI - 04/29/25
- You are in a Reorg Bubble about to Explode - 04/22/25
- The Wrong AI Tool Is As Bad As the Wrong Hire - 04/15/25