Give me the chills

This is a place to share learnings, from technology to life everything can be interesting.

Author: Andrea Grassi

  • Asking for Guidance

    Asking for Guidance

    How do you ask for guidance in today’s world? Here’s what comes to mind for me, based on a recent experience.

    Traits

    I wanted to ask for specific guidance and feedback from some people I admired, people I worked with. I wanted to understand if and how my work, and the way I am, impacted them.

    There are people who are able to spot their strong traits and weak traits easily. I am not one of those people. I often dismiss what I can do or am not aware of what my true skills are.

    Obviously, it is not like I am 100% unaware. But I tend to consider that part normal, just the normal day-to-day stuff that is not extraordinary or special.

    Teaching

    The other thing I wanted to avoid is that sometimes we think we are good at something, but we are actually not good at teaching it or guiding people through it.

    It is something I learned from my singing teacher. She often reminded me that someone might be good at singing, but they might not be good at teaching singing. The mechanism might be different. They might not know how to decompose, or even remember, the pieces that can make someone progress.

    They are so used to being skilled that they do not remember the initial part where you are building the skills. Or maybe they were just talented, so they never arrived at that part.

    Time

    Whenever you are asking people for guidance, you really need to remember that what you are truly asking for is their time to help you.

    Sometimes you might have something to offer back because maybe it is useful for them. But sometimes it might be mostly to help you. To me, that always seemed a little bit unfair.

    The way I do this is to remind myself first: Why do I care about this?

    It is easy to just ask the question, and it is probably shorter, but it is also valuable to explain why this is important. Why am I asking you? Why do I care about this?

    Admiration

    Usually, it is because I admire the person. I highly respect their opinion. I think they are very smart, very empathetic, or whatever. I am asking because I care about something.

    I have always admired people who were able to transform other people through coaching, guidance, and things like that. It is something I have always kept in my mind, even though I surely forget it from time to time.

    Remembering why I care and why I value their opinion is a good way to ask these types of questions, especially when I do not have something to give back to them. They are not randomly chosen.

    The thing I personally give back in scenarios like this is that I share something I probably never quite took the time to explicitly say out loud: I highly value your opinion. I admire you.

    We might admire people, but we might skip saying that because we are worried about power dynamics inside the workplace or about being misinterpreted. Sometimes that is the right thing to do. But I think some people never get the chance to truly appreciate the impact they had on the people they worked with, even in small interactions.

    So saying that out loud, whatever the role, is important to me.

    Credit

    When something makes a difference for me, I always try to reach out to those people, often privately, and just say, “Hey, this is great. I’m super impressed that you’re doing this.”

    The goal is not to have friends or get something back. The goal is really to give credit where credit is due.

    Some people do not have that opportunity because maybe you are too high in the chain of command. But I think everyone should have this kind of guiding light, people being honest with them when they are doing good or doing bad, so they can learn and get feedback.

    AI

    I had writer’s block when asking this question to them because I did not know how to frame the question.
    So I asked for some ideas about which questions I could ask and how to frame them to my dear digital non-friend.

    From those questions, I realized, “Oh, that is the framing that I wanted,” and I brought my framing, my version. The AI did not write the question, nor the content. It helped me framing it and find the right words.

    Doing it this way, you have content that has been informed but not created or augmented by artificial intelligence.

    That is another way to use AI. We always see AI as something that does work for us, and it is powerful in that.

    But it can also be a thinking partner. It can be something you use to have ideas, expand the opportunities, and choose which path and which way you want to go.

    You could still do it with a human, and you would probably get different answers and different ideas. But that is another opportunity to use AI, not for writing, but to inspire your writing and give you some additional ideas that you will finish.

    Endings

    Whatever guidance you ask, expect some people to reply, some to not reply (unless you’re doing it live). Accept whatever you get, and as with always “Take what you need, leave what you don’t”.

  • Slop is Inevitable, Slop is Good

    Slop is Inevitable, Slop is Good

    More than 15 years ago, I decided I wanted to learn singing. Up until that point, I was the guy who didn’t sing. I didn’t sing in public, nor privately.
    We have this image of people singing in the shower. I didn’t sing in the shower.

    Singing was not part of my radar, it was invisible to me, it was a skill I never thought existed, although I could clearly see (and hear) singers. Admittedly, I didn’t listen to that many songs either, my musical culture was limited, but slowly improving as I was being contaminated by different genres.
    And yet, no singing.

    Until one day.
    I don’t know what prompted this need. It wasn’t because I aspired to be a singer, but one day I decided I’d learn how to sing.
    I searched for a teacher and found one in Florence and started taking lessons.

    It wasn’t cheap, and driving there required extra money, so for me this was also a considerable investment.
    But I made my choice, and I wanted to learn.

    I don’t remember how the first lesson was. I do remember some of them, but I know one thing: I was terrible.
    My voice was not precise, my rhythm was out of place.

    That was the beginning. I remember how learning to sing Landing in London (3 Doors Down) felt difficult, although the vocal line was straightforward compared to other genres and songs.

    Still, my voice felt cramped.
    And yet, I continued.

    Many years passed, and I kept slowly improving, but didn’t feel as though I was progressing. During that time, my taste for music changed, especially for voices. I would look for specific skills and abilities from singers because I was able to spot a good one from a bad one.

    I learned about the existence of vibrato. That kind of “bouncing” pulsation some singers were able to do either spontaneously or in a controlled way by moving their belly and their breath. (I didn’t like the singers who did the latter.)

    But I wasn’t good enough for that, and vibrato felt like something out of my reach.

    I continued studying and practicing. The vocal exercises meant that my entire condo would hear my voice practice, me singing various notes and trying to sing songs, often with a lot of passion (even with a lack of skill). I didn’t care, but I’m sure they might have preferred something else to that torture.

    I remember at one point I was about to drop out because I didn’t feel I was getting any good.
    Then two things happened.

    One night, I sang with some friends at a company party and some people actually did like my singing and honestly said they couldn’t believe I could improve that much (again, remember I was terrible in the beginning).

    That gave me confidence.
    But the true unlock was when, one day, I started producing vibrato. Spontaneously. I couldn’t control it initially, it was almost scary.
    And yet, from that day on, it became part of the skill set I had.

    My teacher said that it’s hard to know where we stand in our learning journey. We could be close to the next real unlock, but we’re not aware of it, and that unlock is not visible in front of us, plus we learn incrementally in small steps, so we don’t see this big shift from the inside.

    But from the outside, someone listening to me after years could spot a difference.

    Years later, I decided to learn improv.
    I remember some people were impatient because it was so hard to get good at it.
    The teacher said one thing that still stays with me today: “If you were good at it, you wouldn’t need to learn it.”

    Looking back at everything I’ve learned from scratch, I know that producing something that’s not good, or sloppy, is part of the process.
    Sloppy can be the way you learn. It doesn’t need to be that way forever, but it’s part of that continuous process of learning, self-correction, and evolution that every learning process contains.

    While I always wanted to be talented at singing, I wasn’t. I was determined, passionate, and intentional.
    Even though my results were messy, and my singing was torture in the beginning, that was enough. I was enough.

    That’s a reminder to me that, while it’s easy to be sloppy nowadays, as long as it’s part of a learning curve, as long as that sloppiness is an output of low quality and not an abdication of your work, all is good. You’re learning. You are enough. You will grow.

  • The Problem with Statistics and Averages: The Will Rogers Phenomenon

    The Problem with Statistics and Averages: The Will Rogers Phenomenon

    One of the problems with statistics and numbers is that we trust them as if they’re always trustworthy. But the problem with those numbers is that they depend on which domains and which sources they’re using. Averages are even trickier than that.

    One concept that blew my mind recently is the Will Rogers phenomenon.

    A Simple Example

    Let’s say you’re a sales manager.

    You have two groups of salespeople.

    Group A

    • Person 1 sells 1 item per week.
    • Person 2 sells 2 items per week.
    • Person 3 sells 3 items per week.

    Group B

    • Person 4 sells 4 items per week.
    • Person 5 sells 5 items per week.
    • Person 6 sells 6 items per week.

    Now your boss comes to you and says, “Hey, you need to sell more items per week in both groups.”

    Here’s the interesting part.

    If you move Person 4 from Group B to Group A, you’re moving the lowest performer from the best-performing group into the least-performing group.

    What happens?

    Group A sees an increase in average performance because you’ve added someone who performs better than everyone already there.

    Group B also improves because it lost its lowest performer.

    Both group averages go up, even though nobody actually became better at selling.

    Numbers are tricky

    This example is obvious once you see it, but the same thing happens whenever you have two groups that are separated and you move lower performers from the better-performing group into the worse-performing one.

    That can happen intentionally, or it can happen unintentionally.

    One of the most controversial examples is around cancer.

    Stage 1 is the “entry level” group for when you’re diagnosed with cancer. The cancer is still early-stage and confined to the original site.

    The problem is that 50 years ago we didn’t have the same ability to detect cancer that we have today.

    Some cancers went unnoticed. Some of them probably weren’t even risky, so people simply continued living their lives with a normal life expectancy.

    Over time, though, technology evolved.

    Some of the people who previously belonged to the best-performing group, the people who lived a long and healthy life, are now included in the Stage 1 cancer group because we can detect those cancers today.

    Out of the blue, Stage 1 cancer group appears to have a higher life expectancy, even though there may have been no improvement in treatment, at least as far as this phenomenon is concerned.

    There may, of course, be genuine improvements in treatment, but that’s not the point.
    The point is that the average improved, even though nothing necessarily improved.

    Why? Because people moved between two groups, removing the lowest performers from the better-performing group and adding them to the worse-performing one.

    Every average and every statistic can be tricked.
    Treat numbers with caution, always.

  • Understanding by Asking Questions

    Understanding by Asking Questions

    One of the things that makes understanding easier is asking questions.

    We are often tricked by our own intelligence because we think we understand, whereas in fact we don’t: We’re not truly understanding something.

    What’s stopping us from understanding the solution is getting deeper into the problem.
    But how do you get much deeper into the problem?

    It’s something that I also touched upon in the past.
    I think the point is asking questions, A LOT of them.

    The thing, though, is that you need to go many levels deeper, to truly understand. The first question is rarely enough.

    Questions

    A basic way to approach this is to just ask why.

    For example, let’s say someone invented or designed a solution for a new pen.

    Well, why?

    Maybe they did it because they thought that this new way of writing with this pen is more efficient.

    Why is it more efficient?

    They can tell you that the ink is gliding more.

    Why?

    They can go even deeper and tell you that, to make the ink glide better, they use a new nanoparticle that is not available yet to the market, and that basically creates a new space between the ink and the paper.

    Why? (Or how?)

    They might continue, and continue, and continue.

    Every time you ask why, you go deeper into the problem and deeper into the understanding of the mechanics of the problem itself.

    The mechanics of why

    Questioning helps us building a mental map of the product, the problem or the issue. That’s what makes learning.

    You understand the whys. You understand how things fit together, how they connect together to make something possible.

    From fake ink that glides more, to whatever you have on your plate.

    You don’t need a good question, as you saw: Why and how are more than enough.

    In fact, they might be even better than smart questions, because with smart questions you might risk posing a question only because you think it’s smart, not because it helps you move your understanding forward.

    Whereas why and how are more than enough to get deeper.

    Why and How

    The why is a question that you need to be mindful of, because when you’re directing this question to a human, it can feel judgmental.

    In that case, it’s better to use how and what.

    But, if you’re trying to understand how AI did some work, why is valid and good to use .

    Broad why and how questions are fantastic to get deeper into the problem. I tested this today to understand some work that I was doing with AI, and it proved to be really good.

    You can get much deeper, not only is the AI forced to think about it, but you’re also forced to think about it.

    The more why and how, the better.

    This is a way to get deeper into the problem, to develop an understanding of anything you have around you.

    Don’t be shy, ask why.

  • Cutting Costs / Adding Value

    Cutting Costs / Adding Value

    What is price? It can be a measure to “weight value”.
    For example, there was this person on Twitter that stated stated how $30 per PR shipped is a good price, and I tend to agree.

    The thing we have with AI now is that you can spin up a model ask it to build something.

    Guess what? The model will agree and will start using tokens (spending tokens). There is no pushback. There is no questioning of any assumptions.

    You can build AI skills that manage some level of pushback, but overall, whenever you ask a model to do something, it will do it, sometimes it’ll relentlessy do it.

    This brings everything to another level and raises another question that remains valuable today, a question that was there from the beginning: What’s the value of what I want?

    There will be a time when people will start saying we need to bring the costs down, at that time we will go back to square one because, as I said before, it was never about building. It was always about building the right things.

    This opportunity to directly see the price of an implementation, or at least have a rough idea, gives you an overall understanding that money is a finite source. You want to spend it wisely on things that matter and things that will make a difference.

    But we can’t ignore the fact that we are now in the full bloom of AI. It’s normal to abuse tokens now. People are exploring a new world.

    But, at the same time, I expect that there will be a time when some people will either switch to local models for things where they are unsure of the value, or they will start questioning the product roadmap.

    For example, is it ok spending $1,000 or $2,000 for an idea that might not bring value? These costs will sum up over time and will add up maintenance costs that will make running products even more difficult.

    What we ultimately want is a sustainable product that is easy to maintain and brings true value for customers, that’s the dream right?

    It was never about the price. It was never about what to build or the time to build. It was always about building the right things that solve people’s problems and keep them engaged with your product.

    But this is actually a cool thing about AI because you can now have a rough understanding of how much investment is needed to do anything.

    Before, the situation was slightly more blurry, estimates were difficult, we didn’t have “human tokens”. Now, you are starting to see the cost clearly also (strangely) thanks to measurable token costs which, even though you can’t predict the exact amount of work (and therefore the spending) can help a lot in this kind of planning.

    It will be somewhat easier for entrepreneurs, stakeholders, or product people to think about whether they are spending money in the right direction.

    And, eventually, they will be accountable for how much they spend.

    This is not new: If you want to adopt a new product that costs a significant amount, you obviously need to figure out how make a case for its usage in your company. This will be no different.

    Knowing that what you build is valuable or will make a ROI seems obvious, but now you have kind of a metric to help understanding it.
    Tokens.

  • When Phishing Comes With Your Booking Details and What To Do About It

    When Phishing Comes With Your Booking Details and What To Do About It

    Last Saturday, I received a WhatsApp message.
    The interesting thing was that it contained very precise information about a booking I had with a hotel that I was going to in a few weeks.

    My first reaction was, hey, I need to confirm the reservation. I want to make sure that my reservation is being kept.

    At the same time, I realized that there were a couple of things that were off.
    I quickly realized that this was a very well-crafted phishing message, probably caused by some data leak, either from Booking.com or from the hotel itself.

    I was lucky and experienced enough to not fall for this, but I think it is valuable to share how real this thing can become and what to do when you receive something similar.

    The Message

    Before anything, I’m going through the content of the message. Then I’ll list the actions.

    Looking at the beginning, the first red flag for me was the number, but I didn’t look at it right away.
    Even though I’m in Italy, I tend to frequently have English messages from people and booking agents that write me in English.

    When you read it, you see that in the beginning there’s this number coming from the UK, and there’s this image with “guest support team” written on it, as if it’s coming from a real hotel.

    Then it had something like a message that you would get from the hotel saying, “Hey, [my name]. This message is related to the booking at Hotel X,” which was correct, “from that date to this other date,” and including also the Booking.com correct reference number

    That caught my interest because it was very real.

    The Red Flags

    Then there was this slight twist where they said, “Your booking is currently in a pending state and requires confirmation to remain active,” which was unusual for a normal booking.

    Then I had this deadline: 11 hours and 30 minutes, which is the main thing that every phishing message usually has: some level of urgency.

    You’re getting a message from a trusted source because it has knowledge about you that, theoretically, only the original source should have. Then it puts some urgency: “You have 11 hours to confirm.”

    Then it continued “please review your reservation details in an external website.”
    This was an additional red flag for me.

    The other thing that was strange was that it was a strange website. But that could happen because sometimes hotel booking systems might be on very strange websites, but that could have been possible.

    “Once completed, the reservation will be automatically updated,” the message says.
    If no action is taken, this is the urgency again, the reservation may be released.

    At the end they add the hotel name again plus it is a WhatsApp business account with fake buttons that said “Cancel my reservation.”

    What I Did

    Once I realized that this was fake, I checked the domain whois, I checked the hotel, and I brought everything to the hotel’s attention.

    Instead of acting on whatever I got from this message or replying to the number, I went to the hotel website and to my booking and asked for clarification.

    They confirmed that this was not from them, so someone was impersonating the hotel.

    The real problem was that the attacker had real information about the booking reference, probably because of some data leak either at Booking.com or at the hotel level.

    How To Protect Yourself

    These kinds of attacks are going to become even more sophisticated.

    How do you protect yourself from these attacks?

    First and foremost, don’t trust anything that has some level of urgency in it. Whenever they ask you for an action, confirm it and use the official communication channel.

    Instead of answering a call, call it yourself.
    Instead of replying to a message, send a message yourself to the hotel.

    Don’t trust the calls because the caller ID can be faked in some places. Don’t trust even if they call from the right number. You need to call them yourself.

    You check with them, and if that is confirmed, then you do it when you call them, not the other way around.

    Never trust any message you receive whenever there’s some level of urgency.

    Don’t click links. Clicking links, as I recently discovered, might expose information about you, including that the number was correct. They might also get information about your browser, your IPs, your location, depending on which browser you use.

    You want to avoid interacting with any of these, and the best approach is always to report it.

    Hope it helps.

  • Opinionated vs Not Opinionated

    Opinionated vs Not Opinionated

    You’re maintaining a project that is going to be used by other people, think about a library or a set of components. One of the key elements is to choose whether you want that project to be opinionated or not opinionated.

    Opinionated means that you have a vision, you care about that vision, and you adapt to the goal, so that you get closer to what that vision is. That vision could be a set of values, of ideas, or of constraints that you will place onto anyone using that library.

    On the other hand, if you want it to be non-opinionated, then it means that you’re targeting a wider audience, you’re trying to reach a bigger audience and your project needs to be able to support different activities, different interfaces, or whatever that is.

    The point, though, is that when the project grows, being opinionated could become harder and harder. There are trade-offs between the two ways of working.

    Taking decisions

    When you’re doing non-opinionated, sometimes taking a decision can be hard, because who am I serving?

    It’s kind of building a product, after all: you’re serving a customer type.

    Whereas when you’re opinionated, you’re serving yourself first, or your needs first, and it’s easy to take a stance and say: yes, this goes in, this doesn’t go in. You’re, after all, clear on what should be done and why.

    Keeping the stance

    Now there’s another thing that’s important: when it is opinionated, keeping that stance might be harder than it seems once the project grows.

    The “opinion” belongs to the person who had the vision, and as humans we tend to (or at least that’s the way I see it) create new rules, and think that those new rules are the new opinion.
    But those are just rules, not values, not ideals. They might come out of values as a way to make those values tangible, but that is not a guarantee.

    Being opinionated means, to me, that you want to continue passing down that initial opinion with full intensity, instead of diluting it. You want that concept, use case, ideal or simplicity to be passed down with no exception, so that your product is expressing the need or your idea.

    I think building that type of product is easier when you’re the main user, and it’s harder when you’re looking at it externally, because the more you’re the target customer/user, the easier it is to keep that opinion true to its roots down the line, as opposed to following rules that are synthetic to some level.

    Instead, when you’re building something that is not opinionated and aims to reach a wider audience, you’re facing all the challenges of how should we build this and how should this be implemented.

    That’s something you could compare to products that are wide-ranging, aiming to reach a bigger population, as big as possible, that are generic and broad, and nailing down the user type and the user need becomes the job.

    You’re searching for opinions, opinions that you can serve and include into the use case to advance the product itself.

  • It’s Never the Right Time, Yet It Always Is

    It’s Never the Right Time, Yet It Always Is

    How many times have you thought, “I’d really love to take this trip right now, but I have work, I can’t afford it.”

    Or, “I’d like to have a child, but my life is too messy right now, I wouldn’t know how to handle it.”

    When is the right time?

    When does that mysterious alignment of planets, events, things, and people happen, where everything is in its place and the only choice you have to make is handed to you by what we might call fate, ready to be seized?

    Does such a moment exist?

    A moment where we have the clarity to see that everything is perfect, that it is THE moment, and that only then should we make the choice we need to make: the trip, the children, changing jobs, changing lives, moving away, traveling, getting lost, starting over from scratch.

    I waited for many things in my life. I too waited for there to be a green light in front of the choices I wanted to make, but sometimes that green light never came.

    Just like with having children, I decided to have them even though I didn’t feel ready, even though life wasn’t as it was supposed to be, even though there was something that could have been done better.

    Looking back, I think: I did the right thing.

    The point is that there are two problems behind waiting.
    The first is that it’s not possible for there to be a green light.

    Life doesn’t make room for you; you make room for it.

    The world doesn’t adapt, it doesn’t know what you want. The world around you evolves in ways you cannot predict or imagine.
    So that astral conjunction you desire so much? Forget about it. It will never come.
    It will always be an imperfect choice, with an imperfect outcome, and yet still beautiful.

    The second point is that maybe there are cases where the world does prepare itself for you.

    Where the clearing’s leaves part to let you through
    In those moments, in the frenzy of today’s world, would we even be capable of seeing such a miracle?

    Or would our eyes quickly pass over those displaced leaves, ignoring them? Labeling them as something ordinary, when instead they were part of an immense design that we were incapable of seeing.

    So when is the right time?
    In the end, I think that if you’re asking yourself the question, maybe that is already the right time.

    Because ten years ago you wouldn’t even have asked yourself.
    It wasn’t on your radar.
    But now it is.

    Maybe you won’t be ready, maybe something will be missing, maybe you’ll make a mistake, but the point is that now you have a doubt you didn’t have before.
    And that doubt is the first spark, the first desire that writes in your heart what you want to do.

    You can ignore it and think that you need a whole world around you to justify it, but in reality that small glimmer, that doubt, so frightening and unfailingly beautiful, is enough to justify it.

  • Two Bets on the Future of Software Engineering

    Two Bets on the Future of Software Engineering

    I feel that in the future all code will be written by agents.

    That might not be the time now for every company, but I think that’s where we’re heading. The interesting part is not whether that happens or not. The interesting part is how we get there.

    I think there are multiple ways to reach a point where we can successfully ship production-quality code written by agents. Not in terms of quantity, but in terms of quality. Quality that is comparable to what humans produce today.

    And to understand why that’s difficult, we first need to talk about context.

    Context Is More Than Code

    We all know that context is partly the limit, and even more so the context that is not part of the code.

    Whenever we create a project, we create it with an intention. We define it with taste and goals. Those things are often not written in the code.

    Sometimes they’re documented elsewhere. Sometimes they’re not documented at all.

    The code is only one part of the project. The intentions behind it, the trade-offs, the expectations, and the taste of the people building it are often stored somewhere else, or nowhere at all.

    That’s one of the biggest challenges if we imagine a future where agents write all the code.

    Getting There

    If the world eventually reaches a point where all code is written by agents, there will be multiple stages to get there.

    Right now, what we’re seeing is that some people and companies are using agents directly, while other companies are all-in on using AI to augment a single developer.

    Those are two different ways of approaching the problem.

    I think the first group is going to have more trouble in the beginning because, again, the agent doesn’t know everything. It doesn’t know the taste, the goals, or the requirements that exist outside the immediate scope of the task.

    That gets messy quickly.

    All-In on Agents

    This is similar to what OpenAI described in their Harness Engineering document around projects that are largely AI-generated where the agent does all the work, 100% code is written by it and it’s async. It doesn’t need a human to start working.

    That poses some interesting questions.
    You want to review the code. You want to make sure it’s clean, that it doesn’t introduce duplications, and so on.

    At the same time, you want some kind of ownership of that code by the AI itself.
    So you need to build processes that revolve around making sure the code continues to evolve asynchronously with minimal human interaction.

    Humans are still involved, but they’re not guiding every step. They’re reviewing, controlling behavior, and making adjustments.

    The goal becomes enabling the system to continue operating without needing a human to constantly push it forward.

    AI as Augmentation

    The other group is probably moving slower, but (probably) at higher quality.

    They’re controlling the agents more. They’re augmenting their workload rather than delegating ownership completely.

    But at the same time, whenever we reach a point where we can safely trust agents to do their work, or at least part of it, they won’t have learned how to guide agents in that way.

    Current harness engineering and context engineering revolve around addressing the limitations of today’s agents, but whenever a new model comes out, you’re suddenly facing a different set of problems.

    Some of the techniques you’re using become useless, for example models already require much less prompt engineering than they did before.

    Trade-offs

    I believe some companies are spending time on things that won’t matter anymore in a few years or maybe months.

    One group is optimizing for quality and control.

    The other group is optimizing for amplification and learning how to operate with autonomous systems. It’s a trade off one has to make.

    The Winner

    In the long term, I think the first group, the one adopting AI-written code more aggressively, is going to be faster, assuming they can sustain the errors long enough (or have a quick feedback loop for humans and AI agents) for the models to improve.

    It’s a bet, of course. I’m betting too as I can’t predict the future.

    My take is that models will reach a point where they can remove most of the issues they created. But to get there, you need to enable each model and agent to self-serve and self-direct.

    Even something as simple as checking that reporting is consistent could become a task handled by an agent.
    Or maybe it’s something that won’t matter anymore because the interface itself changes.

    Maybe software stops being primarily UI-based and becomes something else, who knows.

    Everyone Is Betting

    I think everyone is betting on something right now.

    The companies going all-in on agents are making one bet.

    The companies focusing on augmentation are making another.

    I think that learning how to give agents work and let them operate autonomously will be important for companies right now, the more that is possible, the better positioned your company will be if that future arrives.

    What will you be betting on?

  • Nobody knows the AI Future

    Nobody knows the AI Future

    Recently the CEO of ClickUp announced a layoff of twenty-two percent, and that was also bundled with a lot of ideas for the future, about restructuring, and about what it means for now.

    Here are my takeaways.

    1 – AI is shifting everything

      Nobody knows what the future will look like. Don’t trust who says otherwise, because we’re still in a moment in time where little do we know how AI will shape the world and how we will adapt to it.

      We’re currently doing our best, as usual, but the world five years from now will probably be quite different from what we have today.

      2 – Tokens are a limiting factor

        Tokens are a limiting factor, and inefficient tokens are a hurdle and a cost.

        This is not fully shared by the CEO. He mostly talked about letting people use unlimited tokens, and I agree with the direction. Token usage can absolutely be a limiting factor.

        Right now you want to leverage as much power as possible that AI can give you, because you can expect some of the services will bump the prices up, and we’re already seeing this.

        3 – Critical thinking matters even more now

          This is my take, but you want to build even more critical thinking as you hand off more work to AI.

          If you didn’t have enough critical thinking before, now is the time.

          If you had it before, be aware that you’re trusting AI more than you would trust a human. You need to address that.

          4 – Bottlenecks will define the future

            Everything that slows AI adoption or AI distribution is currently a bottleneck.

            It’s not about “AI can code.”
            The question is: assume that everything is done by AI, what can’t AI do currently?

            Those are the bottlenecks you need to figure out.

            Again, this is mostly a vision for the future, so it might be wrong.

            5 – Deprecating old systems

              If you want to shift the mindset entirely, adapting might take longer. You want to reinvent because adapting them might require so many different changes and what you want is speed.