The tech industry loves to talk about “10x”. Many claim they’re much more productive thanks to AI tools.
But have you? If you don’t, do you think these people are exaggerating, or are you missing the breakthrough?
Here are some thoughts about how to unblock your company and data team to achieve the productivity gains you want.
Is your workflow AI-native?
You need to put yourself in the right environment first. Development workflows from 2020 weren’t designed benefit the most from agents, so you need to seek out how to adjust.
Think about some of the stages of codegen:
Stage 1: One-off questions
This is human driving code, use for one-off questions. It only has the context you give it, and then you have to copy/paste back into your IDE.
Stage 2: Autocomplete
This is human driving code, IDE has tab autocomplete. I skipped this step, finding it too distracting (often wrong) and not wanting to feel lazy. It was too focused on predicting what comes next and not planning what you actually need.
Stage 3: Task-level
This is human assigns task, agent attempts to code it. For a lot of core project work, this can be fine. Requires you to do scoping and good job of describing what you need. This + the validation can be sneaky in bringing total amount of time closer to the time it takes to just do it yourself.
Stage 4: Project-level
This is human assigns project or story, agent designs and codes it. This is where you really start applying the leverage, since you can extract yourself from the IDE if you choose. That means less context switching and you can keep the project manager hat on.
Stage 5: Owner-level
This is agent decides what projects are needed, then designs and codes them. I haven’t gotten here yet, let me know if you have and what issues came with it beyond the predictable.
You don’t need to get to Stage 5, and I probably wouldn’t recommend it. Somewhere between Stages 3-4 seems right, so if you’re in Stage 1 or 2, this may be a sign to experiment.
Bottlenecks to increased productivity with AI
Writing code is cheaper now, but it probably wasn’t your biggest bottleneck. But we all know that’s not the entire job. To maximize your gains, you’ll need a strategy for these 3 issues:
Stakeholder alignment
You want to change your attribution logic. You can write the code faster now than ever before. Is your organization ready for that speed? Or do you need to set up meetings, discuss, validate the expected differences between existing and proposed methodology, discuss again, and then maybe punt until next quarter?
For things that are not contained to just the data team, especially logic that stakeholders in the business have a serious interest in, you probably can’t expect to fast-forward through the legwork.
Change management and maintainability
You don’t want to turn your codebase into the Ship of Theseus, rewriting everything on a dime just because you can. You want to make sure that you’re making changes with good reasons — as every time you merge a PR, you’re increasing the chance of silent failures you may not catch til much later.
Sometimes friction is a good thing.
A more disciplined codebase means you’ll move faster and more accurately on the tasks that do matter.
The last mile problem
Sure, you can get to a draft much faster. But then how long does it take to get to a finished product? When you’re not writing the code yourself, there’s more surface area for silly bugs and you need to read through potentially a significant amount of code to spot them.
The silver lining? If your coding agent is connected to your data warehouse, you can spin up several data validation queries in seconds to confirm your numbers are correct.
Conclusion
Is 10x productivity from AI a myth? At the individual level, no. Assuming you have the right type of work and can work more AI-natively, you absolutely can move much faster.
But for teams and businesses, it’s harder. There are many more obstacles and blockers in the way. It is the job of founders and executives to clear the path.