Photo by Alvan Nee on Unsplash
First of all, sorry for the long hiatus. Things have been tricky in the personal department, which basically means trying to keep some plates spinning: family, work and, if possible, some time for myself (that last one I usually fail).
And as you know, three months in the tech industry might as well be three years. I find myself trying to catch up every time with which models are new, but also… wait, now we have an “assistant” that looks like Flubber? (If you are old like me and got that reference, I salute you.)
Then we had a black October with all the crazy vulnerabilities:
Plus some honorable mentions from outside October, because apparently the universe does not care about my calendar:
How about the math “breakthroughs”? I don’t get it. Is P not equal to NP, or is it?
I’m getting confused. Alexa, play Despacito and tell me if my Docker container running ArchiveBox is also vulnerable. It seems that everything is getting vulnerabilities these days, including people:
And I’m not even taking into consideration all the rogue/ninja stuff from the other big boys, like:
And so on…
Anyhoooowwww…
I just wanted to touch base with all of you and have a small “ra”. A “ra” is like a rant, but not fully. Call it a semi-rant: conceptually the same as a rant, but with a lot of questions without answers and just a small light (from my perspective) at the end of the tunnel. For this, I’m going to give you a view of the current situation I see some friends facing, and some of my own, and if possible my takes, without flying too close to the sun so my current employer doesn’t feel the need to have a “fruitful conversation” with me about my position in the company.
Everyone can PR
Lately I’m facing a situation where the demand for AI usage has grown exponentially, and without trying to sound like an asshole, the phrase “you can’t make a baby in one month with nine moms” is, from the perspective of certain people, “not true” anymore. Why? Well, you know why: that beautiful transformer, that autocomplete on steroids…
For some reason, more people are starting to think that you can brute force more and more stuff and still expect the quality, and the eyes needed, for things not to break. And that doesn’t even include the near-impossible situation of having someone who actually understands what is going on inside a codebase: the architecture, and overall what the project is built upon.
Call it PR love bombing: you get showered with agent-generated pull requests as proof of how much everyone loves velocity, and you are expected to feel grateful instead of asking who is going to review them.
What happens when you are already using agents with the right set of skills, rules and workflows, taking advantage of worktrees, subagents, etc., etc., and STILL you get marked as “slow” under the eyes of Olympus?
What happens when your lead, or someone even higher, feels like asking their agent to check a ticket, create a feature or implement a fix and send a PR can be done in a few minutes? That gives the person who typed the prompt the false sense that they can do what other people, on other teams, are in charge of and responsible for.
Why ask for guarantees on something that is still in the process of adoption and has no clear organization or rules in place?
This reminds me of a phrase a designer friend of mine told me once: “Sometimes my client just tells me, ‘this needs more design’. He has no idea what that means, but basically he is telling me to read his mind and figure it out, and that is why I’m getting paid.”
Well, I feel that these days we are getting something similar: “this just needs more AI”.
Medicine has a word for this: iatrogenesis, harm caused by the treatment itself. The doctor means well, the pill is real, and the patient still ends up worse than before. Every “helpful” PR nobody read before opening it is a small dose. One is fine. Twenty a week and the team is sick from the cure, not the disease.
Overall, it seems that the nature of many leaders is to sell the concept of being more efficient, innovative and AI-driven each year, which, as you can imagine, creates a “good halo” for the business.
Now, before I get lynched by management, I want to be clear about something: I do use and like LLMs. I find the evolution of this career something interesting to learn from, to grow from, and I want to be part of it. BUT, just as abstractions took over the older languages, new architectures took over the old ones, and new hardware replaced what came before:
We cannot deny that there is something to discuss and take into consideration when we want to add AI to a system or workflow that already has AI in it. It is not about a change of paradigm; Moore’s Law already told us, back in 1965, the changes that were coming. But in the process of trying to adopt, every quarter, sprint and day, a better usage of the tools at hand, and of understanding where to be “better”, we are getting rammed not only by hacks, exploits, vulnerabilities and spies (it seems), but by the blind eye that takes everything to the cliff just because we need to get there faster.
I would be wrong not to understand the importance of the race a company faces to stay in business. But what happens when the people who want to help you win that race are not only already putting the foot all the way down, basically adopting the “all gas, no brakes” concept, but are also showing you the cliff, and there is still a demand for “this needs more”? Is there a balance that works for everyone?
I really hope so.
And it is not only me being grumpy. Harvard Business Review published a piece last year on what they call “workslop”: AI-generated work that looks like good work but lacks the substance to actually move the task forward, so whoever receives it has to clean it up. Four out of ten workers had received some in the previous month, and each piece cost them close to two hours to deal with. Their follow-up from this June goes one step further: the slop does not stay in the inbox, it mucks up the processes of the whole company. Sounds familiar? A PR nobody read before opening is just workslop with a green checkmark.
Anyway, thanks for reading, and sorry again for the overdue entry. Hopefully I don’t take too long to write about some actual engineering, but as the title of this blog says, sometimes this will be about tech, sometimes it will be about dogs, and some other times about something that hopefully makes sense to someone and ideally makes you laugh a little. BIG HUG for you all!