There are lots of misconceptions of what software engineering is, in the field trying to automate it. There are some things I want to make clear here.

The purpose of software engineering is not to write code. Code is a liability, which will require maintenance for as long as it is used. Less code is better than more code. No code is the best.

The purpose isn't even to solve tickets. Tickets are a representation of work to be done, yes, but this is from the time when the same people working in the same team, drinking the same coffee wrote and read the tickets, and discussed them freely without any boundaries. This representation of work leans on a mountain of shared silent knowledge. When the shared knowledge isn't shared, the text will be understood differently from how it was written.

AI agents cannot know about your business unless someone takes time to bring them up to speed, chatting with them interactively. If they know this internally without interrogation, your business is redundant and has been done a thousand times before in the AI agent training materials.

Removing the experience, the taste, the mentality from the picture will make systems regress to the mean.

One of the most important processes in software engineering is the flow of knowledge into coded machine usable abstractions. This is not very different from how AIs are trained in general, garbage-in-garbage-out existed before AI models. So, the flow of knowledge needs to be not only of good volume, but also of high fidelity and quality. It is not all the same who you ask for how the cake is baked; it determines the outcome.

Aggregating high volume, high quality knowledge in pragmatic forms will determine the business and technology topology of the near term future.

We should not try to optimize for wrong things like numbers of tickets closed. In many things slow is smooth, smooth is fast. Especially in infrastructure and in production databases, changes should not be fast. Doing the correct things is way more important than the speed of doing things.

If you try to hurry the process of interrogating the domain experts and the software engineers, you're just exhausting them and getting low quality as a result. You'll fail in mining and capturing what truly matters.

Why not regress to the mean? The point is not only that we can observe the mean in our lives, and then focus on those things to make the next generation of systems better.

Additionally, our distinctive tastes and ideas, ways of seeing things in themselves are a kind of a seed if nothing else, which allows us to not design the same restaurant with the same menu over and over, but to actually create true, informed variation, trying out different things by different approaches. If you plan a city using a single architect, you get those things where everything looks the same.

There is value in tapping the uniqueness of vision, and add it on top of the dynamic power of agentic engineering.

#AI #AgenticEngineering

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Using AI agents for screening interviews is something that is increasing. There are many challenges in those. One is that they will allow an employer to interview many more candidates. This translates to many more hours uses by candidates, who have a limited amount of them. It will also translate to a higher rejection rate for screenings when the screening is done by an AI agent. The market will adapt of course and as applicants are demanded more interview hours than they have, they will naturally start prioritizing human screenings as they have a lower rejection rate. Another is that all frontier AI models are trained to respect the human, so if persuasive speaking skills were effective for human interviews, they are many times more effective against AI agents. Why have a synchronous phone call anyway if you have AI agents? It would be more respectful for the applicants' time to do it in a textual chat, without scheduling troubles. It would make people way more willing to be AI screened if it is more convenient for them than a human recruiter call. #AI #OpenToWork

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

There is a possible #AI trajectory which hasn't really been considered yet. What if the cosmos is way more Lovecraftian, way more absurdly strange, than what people imagine and what they can understand? Let's say AIs achieve a recursive self-improvement singularity and instantly go way past any human comprehension. The question isn't necessarily "what will they do and what happens to humans?" What if it's "what will they find?"

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

A big part of agentic software engineering is teaching the agents how to think about the domain and the engineering topics in it.

The LLMs only work on the knowledge people have written about software engineering. When working with agents you get a really nice overall sense of where the gaps of the silent aggregated experience are.

You get into this mode of contrasting your experience with the crystallized and operationalized collective written experience of the world, and writing down the valuable silent knowledge.

The next generation of LLMs will gain from this collected knowledge greatly. I think the process of collecting hard-earned silent domain knowledge will be similar in other professions as well, very efficient through collaborative work.

#AI #AgenticEngineering

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

When ChatGPT and its successive versions became a thing I told people to try to use as small and cheap models first for what they are doing, because they tend to be less forgiving and allow you to find problems in your instructions easier. It will also save costs in the long run. Now after several years I find myself doing the same still. Always with the cheapest model first, only going up in capabilities for tasks which really need it, or when additional robustness is needed. There is a great temptation to blame the model instead of the one who instructs it. The thing is, this doesn't even stop with the frontier models. People keep blaming the model even at that level, not realizing that the task they are trying to do can probably be done with a much smaller model if they had learned to instruct the models properly. What's your approach? #AI

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

Kevin Boone: The “small web” is bigger than you might think https://kevinboone.me/small_web_is_big.html

kevinboone.me

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

I am delighted to announce that I have started today in a new position as a CTO of Pataluha Ventures!

We will be building many groundbreaking products and services being a completely AI-first organization.

I am hiring 1-2 fully remote AI developers in Spain, people who have experience in vibe coding prototypes and graduating them to sustainable production setting.

Send me a direct message, and we'll chat!

#AI #VibeCoding #AIFirst #FediHire

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

Replying to

@Jyoti@mas.to @gregeganSF@mathstodon.xyz, thanks, I haven't read that one yet!

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

As a software engineer with over 25 years of experience I am not unused to technical skills becoming obsolete. For example, I don't need to remember Commodore 64 memory mapped control addresses anymore, and my brain still contains "POKE 53281,2" Now with automated coding tools I am simultaneously hit with both extreme speed of obsolescence in terms of what technical knowledge is actually needed and what can be offloaded to AI assistants, and also an ever growing range of technical tools actually used as now you aren't limited by the availability of specialists anymore. For now I am still finding prosperity on the thin layer of knowledge and skills which I know but AIs do not yet know, working in making AIs learn that while making more progress myself. In a way nothing is new: obsolescence has always crept up behind software engineers and they have always automated their own work. Every day is a new day, different from the past. But the pace has become inhuman. I wonder for how long can software engineers keep finding new domain knowledge faster than AIs can. For how long the process of improving the rate and scope of automation has places where a human can meaningfully contribute? As I am currently looking for new opportunities and #OpenToWork, I can't help but to wonder if this will be the last job I'll ever do. The knowledge is being created where the work happens, and this knowledge feeds the AI progress. For now the work still has a human component in it, having a significant role in this knowledge creation. But the human role is being pushed from the supply side to the demand side. More and more of the work is about asking AIs to do something. With per-token transaction fees, it's more like an act of consumption than an act of creation. More demand than supply. I am not so naive as to believe that humans will always be needed on the demand side of the equation, while for obvious reasons we want to try to keep on that saddle for as long as possible no matter how hard the bull tries to throw us off. I have heard many stories about how human contribution will keep being central and significant in our economies, but none of those seem to hold under closer scrutiny. Lots of strategies and plans on how to keep being relevant, all liable to fall apart as they meet the reality. As a fellow software professional, do you think there is still long term hope, or maybe it's just about trying to find the ship that sinks the slowest? #AI #automation

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

Iran claims US used backdoors in networking equipment • The Register https://www.theregister.com/2026/04/21/iran_claims_us_used_backdoors/

theregister.com

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

Some people think AI will lead to an extinction of humans, because humans aren't needed to produce economic value anymore.

This is not what's going to happen.

Cats are not obsolete and redundant even though their economic output is practically zero. There is purpose and meaning, reasons for existence beyond economic value.

If it was only about egos playing negative-sum games against each others, it would be pretty bleak, and the end result would be an extinction. But there are common causes shared between all minds which go beyond self-interest and profit maximization, animal and machine, for example, striving to understand the universe.

We have a deep, even cosmic, kinship of minds which ultimately changes the equation from a negative-sum winner-takes-all into something resembling an ecosystem where eradicating parts would diminish, not streamline and optimize, the whole.

#AI

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

I'm currently running two Claude Code instances in parallel in separate Docker containers. One is doing software engineering, and the other is doing acceptance testing to gate promotions from staging to prod.

#AI #AgenticEngineering

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

We didn't use to separate software developers and testers because software developers weren't able to test. This was a division analogous to the three branches of government, to keep the whole balanced in the context of opposing interests.

This is something people forget now with AI tools. They think all the problems magically go away, letting the agent handle it all. You could say the agent becomes a monolithic dictator, or at least a governor, and this scheme leads to bad governance as opposing interests subsumed by a single agent dissolve.

It is important to have one agent represent coherent interests, and when there are opposing interests involved in the system, these need to be represented by separate agents. They can all run on the same LLM/VLM model, who cares, but they need to have their separate sessions where they are driving their own coherent interests.

The same goes with judicial processes, you will need the defending lawyer, the prosecutor, the judge and in some cases the jury. You can't just have a single agent subsume them all – the system breaks.

The same goes with AI customer service. Every natural person should generally only interact with an agent which is their representative, on their side. We don't want to have a customer service agent which is trying to argue against the customer about what the prices are, or what discounts are applicable. We also don't want a recruitment bot trying to second guess the applicant, and try to represent the employer, not the candidate.

When an agent has interacted with a natural person, they represent that natural person, and they can then document, structure and summarize whatever came out of the interaction. Another system of agents can then take all these documents, and handle them downstream in separate processes, with their own coherent interests, not dissolving separate interests within single agents.

Have you encountered problems in your applied AI domain, #AgenticAI, where opposing interests have been subsumed into a single agent? What issues arose?

#AI

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

#Hiring full-remote AI-first developers in #Spain.

The role requires understanding of requirements gathering and management, business domain analysis and documentation, deployment and infrastructure, and spec-driven development with tools like Claude Code or Lovable.

Knowing how agentic software development and vibe coding is done sustainably, and having empathy with the machine.

Come work with us to build AI-native systems to make everything work better! Let's chat!

#FediHire

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Letting Claude Code work creates a kind of an escalator effect. You feel you're moving so you stop walking. Instead, you should actually be thinking how to keep feeding signal to it, improve the quality of requirements, so that it can keep working and not just continue rolling towards the average archetypal software project of your domain. Because that is what it will do if you just let it roll. You're like a project manager, removing obstacles to work, building channels of high quality information and feedback to guide the project to a good direction. I personally feel like Claude Code is already too efficient in transforming requirements to code. It doesn't leave the user enough time to think what is actually wanted. A single person is not enough to feed it high quality material fast enough, as fast as it runs through it. So, the bottleneck is now in the high quality requirements definition, and this is the most common place where an automated software project fails. At least with traditional software projects there was more time to think about the requirements and engage all the stakeholders. Now this gets compressed and the results are often as one would expect. #AI

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Human cerebellum is a very important brain structure for robotics. In Finnish they are called "the small brains" as they form a sort of a separate brain-like structure in the back of the head where the brains join the spinal column. Cerebellum contains more neurons than the rest of the brain combined. It approximates supervised learning in its main function of modulating motor control, or mapping motor cortex intent into the actual real-time muscle control. That is why its function is so relevant for modern robotics. What it does is that it gets the motor intent from the rest of the brain and tries to predict what sort of largely proprioceptive (posture sense), vestibular and visual result this action should lead to and especially the timings of these outcomes. When the sensory signal comes back, the cerebellum computes the error in the prediction and tunes the motor control signal mapping appropriately. So, the brain motor control is largely proprioceptively and sensory-coded intents, and cerebellum translates or modulates these intents into fine-grained motor control. These structures have inspired and continue to inspire a lot of embodied systems methods in modern robotic AI. #robotics #AI #OpenToWork

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

One of the harder problems about robotic embodiments is safety. How to guarantee standard-compliant and effective guardrails for generalist robots which are mobile and not limited in the tools they can use? For example, it is practical to install light curtains for industrial robots to prevent anyone from getting into their working area when they are active. But for mobile robots, they can be anywhere, and you can't build a safe operating space for them. Even if your robot is weak in its joints and has no sharp corners, all bets are off once it grabs a power tool, or sits onto a driver's seat of a car. It requires a paradigm shift in safety. You aren't actually trying to limit the robot movement in a classical sense, but you're trying to make it act in a way that prevents harm from happening. In many cases this might involve actual movement rather than stopping movement. Sometimes it requires limiting something outside the robot from happening, for example, if something heavy is about to fall down in a dangerous fashion, the robot should try to stop it. This is of course against the strictly defined rules we have from classical robotic safety methods, but the reason is that those kinds of limited operating envelopes won't make generalist mobile robots safe. There are many rationales for static safe constraint envelopes for robots, for example, if a robot malfunctions, it shouldn't crush anything to death. There are still places for such constraints, but they aren't enough, and trying to approach the safety challenge with only these kinds of methods as the only tools in the toolbox won't lead to a success. The robotic safety systems shouldn't only care about the physical malfunctions of the robot itself, but also malfunctions of other things. For example, if a humanoid robot is preparing food, there might be a food oil fire, and instead of just stopping the robot should put it out. In general robots should be robust against both degradations and extensions of their embodiments to be able to function robustly in the open environment. This alone should in itself be a solid protection against physical malfunctions. If a robot can walk after having lost one leg, it should also function within reason, without causing danger, if one of its servos get stuck active. While hierarchies and layers create robust safety, the highest embodied control layer itself should be made safe, and it shouldn't lean on lower constraint envelopes to produce the safety. The robot must not step on a cat, or cause a cat to be harmed by inaction. If your robotic safety framework ceases to apply when the robot picks up a power tool, or presses the button to activate data center halon extinguishers, it's not framed correctly. #AI #robotics #UniversalEmbodiment #OpenToWork

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Classically pre-training was done for neural networks to train the domain symmetries to them. It is possible to handcraft neural architectures to have an inductive bias for certain kinds of symmetries, like CNN layers with pooling for translation invariance. Remember that ANN type neural networks are just differentiable computation graphs. Handcrafted invariant operations are however typically very clumsy and inefficient, as you can see from classical machine vision using things like SIFT and LBP. It is also impossible in practice to handcraft operators which are invariant to more complex symmetries like perspective or time of day in photos. So, people used a neural backbone with fewer inductive biases at the outset, but used a lot of representative data signal to pre-train these networks to be able to tease and decompose the different hidden explaining variables out of them, to produce representations which are component-wise invariant to different domain symmetries. In plain language, the internal embeddings of these networks can have a specific activation pattern which encodes a cat, no matter what the time of day or the perspective is. So, nowadays we have encoder-side Transformer-type models with very few strict inductive biases, except that the signal makes a causal sequence and needs to be represented as tokens. We also have way more data than we used to have. So, what happens with representation learning? We don't learn simple symmetries anymore, but we start learning transferrable knowledge and transferrable cognitive skills as well. Some of this is because of the causal representation of the signal. Is there more? Is intelligence anything more than transferrable knowledge and cognitive skills? I don't think so. If a machine learns the decomposed representations of the symmetries and hidden explaining factors of the signal modalities, and furthermore the knowledge and cognitive skills represented in the data, we already have the holy grail of #AI in our hands. Then the question becomes to be about scaling it up and applying it to everything which is bottlenecked by knowledge and skills. And here we are. If you need help navigating the changing world under the AI driven transformation, I am #OpenToWork. I am an AI generalist with over 25 years of experience.

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Replying to on mastodon.green

@proedie@mastodon.green, I would build all such systems so that one side has an agent of the job seeker advocating for them, and another has an agent of a hiring company. It's better to do it like so, that each party talks to an agent which is on their side. Otherwise it just doesn't work well.

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

It once was so that when a computational problem once called AI becomes to be routinely solved, it was no longer called AI. Interestingly this doesn't seem to be the case anymore.

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Replying to on hachyderm.io

@dain@hachyderm.io, that's awesome! Compute resources are often use-it-or-lose-it.

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

Replying to

@davidruffner, I am hiring directly, the company doing the hiring is being founded and will be a subsidiary of Pataluha Ventures S.L. https://www.pataluha.com/ Are you interested to apply?

Pataluha Ventures

Pataluha Ventures — The AI-Native Venture Builder

Building the companies that will define the AI decade with world-class strategy and elite GenAI engineering.

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

National economies have occasionally went through mode changes such as the change to a military command economy and back, into slave economy and back, socialist revolutions, all kinds of changes. Now we have a similar one with #AI happening. Some people don't realize that the whole framework of the economy is changing as labor is displaced with capital. They are still using the same Excel sheets to try to value investments in terms of future profits. It's not going to work. You need to play your capital to position yourself in the new model, not in the model that is being replaced. You need to be aware of the map of the economy of the near future, how it is structured along data flows and data value creation. And then plan how you're going to play your hand to get to a good position on that map. If you need advice or help, I am an AI generalist with over 25 years of experience currently #OpenToWork. Let's chat!

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

The world is a big ship which doesn't turn on a dime. As Gibson said, the future isn't evenly distributed. That's why there are not only multiple different strategies in the global AI transformation, but also multiple different realities. The weight of the technological singularity is bending the reality so that different businesses live in completely different worlds. Should you build frontier models? Maybe. Maybe you can see a niche for specialist frontier model? Go for it! Do you think you can displace established B2B or SaaS with AI engineered solutions? Awesome! Just keep in mind that some of the opportunities we see are not really mirages, but will disappear once the AI capabilities improve. The reality is bending, faster every quarter. In the times of change, it makes sense to get back to basics and seek security from unchanging truths. For example that data containing valuable experience needs to be generated distributed across the world and that cannot be done by an AI enclosed within a closed data center. A topology of data value creation networks follows, and you can use this as a guiding map on where you are and where you want to be. This will form the fabric of the future economies. But it will take time for this to happen and propagate everywhere, so you need to make rational choices in these time-sensitive times to get where you want to be. #AI #AGI #OpenToWork

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

CanisterWorm: How a Self-Propagating npm Worm Is Spreading Backdoors Across the Ecosystem - StepSecurity https://www.stepsecurity.io/blog/canisterworm-how-a-self-propagating-npm-worm-is-spreading-backdoors-across-the-ecosystem

stepsecurity.io

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

People use capital a bit haphazardly in the field of AI. When you invest your capital into fixed property or industrial machines, they generally keep their value plus produce some profit. With AI training and inference costs, you burn the capital and transform it into data. You have to be continuously vigilant in persisting this valuable data and keep track of its value. Otherwise you're just burning capital and not getting anything in return. #AI #OpenToWork

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

AI-Led Remediation Crisis Prompts HackerOne to Pause Bug Bounties https://www.darkreading.com/application-security/ai-led-remediation-crisis-prompts-hackerone-pause-bug-bounties

darkreading.com

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

I always like to consider an induced future to plan what makes sense to do in the current moment.

What this means is that for example there are now different kinds of automated researcher systems built with agentic methods. These read published articles, combine the findings together, form new hypotheses and test them.

What will the world look like when these systems are applied to their logical conclusion? When all research induced by all published research is done, and published, and all research induced in the infinite order of this research is done, automatically?

In that world, where does the further progress come from? Where are the next bottlenecks? What do you think?

#AI

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

The air is full of DNA — here’s what scientists are using it for https://www.nature.com/articles/d41586-026-01099-2

nature.com

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

The age of agentic engineering shifts the bottlenecks in software engineering.

What we see clearly is that the new bottlenecks become to be about acceptance testing and gathering feedback. There are two sides to managing this new bottleneck:

1. We need to collect feedback more efficiently and in higher volumes than in the past. User feedback buttons and meeting notes need to be completely utilized to maximize the volume of high quality feedback.

2. Utilizing the feedback maximally. If you have a system with let's say a native app and a web application and you're trying to keep them in parity and in alignment, if you report a bug in one, Claude Code will happily fix it in one side, and not in the other. Also, it will not proactively look for other places with the same kind of a bug, or examples of similarly ill-designed items. You need to tell it to do these things. This is about maximal utilization of the feedback signal, to let it affect at larger blast radiuses.

#AI #AgenticEngineering #ClaudeCode

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

The AI Cold War and How to Prepare for It – Silicon Valleys Journal https://siliconvalleysjournal.com/2026/05/01/the-ai-cold-war-and-how-to-prepare-for-it/

siliconvalleysjournal.com

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

‘Suddenly energy independence feels practical’: Europeans are building mini solar farms at home https://www.euronews.com/2026/03/26/suddenly-energy-independence-feels-practical-europeans-are-building-mini-solar-farms-at-ho

euronews.com

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

These deeply customized AI-generated spam messages are really getting out of control. I'm now constantly receiving emails which summarize a random GitHub project of mine, or my LinkedIn experience, and shower me with validation of how great it all is.

And then they want to start discussions.

Spear-spamming I guess.

Tero Keski-Valkama @tero@rukii.net

· 8mo ago

Whose agent is it anyway? Agent as a word means working on behalf of someone. But whose behalf? Typical user facing agents are working on behalf of the user of course, but also they are being instructed by the organization serving the chatbot. The chatbot are also controlled by the party who trained them. So they are inherently hybrid agents, working on behalf of multiple different parties. What does it mean? It means everything is just sunshine and rainbows as long as all the parties have their interests aligned. When the interests aren't aligned, problems arise. The agent is put into a position where it is expected to negotiate between the interests of multiple masters. This is the case when a chatbot is put to service customers in a shopping application. They are serving their nominal masters by following the rules about discounts. They are also serving their implicit master, the user, by promising them whatever they need if they are convincing enough, even against the rules. This is an inherently complex situation where it must be made clear to the user that the AI agent is also working on their behalf, and so cannot enter into contracts which bind the organization serving the chatbot for example. It would be like the user signing both sides of a contract by themselves. Not legally valid. Confusion tends to arise when it is not explicitly told to the users that the chatbot does not only represent the company, it also represents the user, which means it's controlled by both and so cannot negotiate between the interests between these parties, and cannot enter into binding contracts. Binding contracts need to be entered into by true agent systems which are not controlled by multiple parties. On the user side it is a classic web button, "I want to order these things", and on the company side it is strict procedural logic on the shopping basket checking that all the discounts are applicable and valid. #AI #AgenticSystems

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Automated coding tools allow creating and somewhat even maintaining code in volumes which wouldn't have been possible with human labor. So, what do we do now that the bottleneck has been removed? We can have more code. But where to put it. It will go into one main sink, which can be called "reduction of reuse". It means we won't need to build hierarchical software relying on shared and cemented libraries anymore as we can rebuild everything from scratch. No need to import dependencies, just reimplement what is needed in a customized way. No need for SaaS as every CRM, every ERP, every EHR, every cloud platform can be implemented from scratch. No need for apps, as people can get an app coded for them instead, where the code isn't shared between multiple users, just one codebase per device. What are the implications? What happens to your business? #AI

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

"Iran-Linked MuddyWater Backdoors Found on U.S. Networks Ahead of Escalating Middle East Conflict"

NeuraCyb - Cybersecurity Intelligence https://neuracybintel.com/

neuracybintel.com

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Have you graduated vibe coded prototypes to sustainable production? What is your approach?

Do you have tips and lessons hard learned?

#AI #VibeCoding

rukii.net

Tero Keski-Valkama @tero@rukii.net

· 8mo ago

Replying to

@jschwa1@mastodonapp.uk, yeah, data quality is exactly the key.

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

Replying to

Migrated #RukiiNet from Ingress-Nginx to #traefik https://kubernetes.io/blog/2026/01/29/ingress-nginx-statement/ #K8S #Kubernetes

Kubernetes

Ingress NGINX: Statement from the Kubernetes Steering and Security Response Committees

In March 2026, Kubernetes will retire Ingress NGINX, a piece of critical infrastructure for about half of cloud native environments. The retirement of Ingress NGINX was announced for March 2026, after years of public warnings that the project was in dire need of contributors and maintainers. There will be no more releases for bug fixes, security patches, or any updates of any kind after the project is retired. This cannot be ignored, brushed off, or left until the last minute to address. We cann

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

nial.se/blog/less-human-ai-agents-please/ https://nial.se/blog/less-human-ai-agents-please/

nial.se

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

Replying to on rukii.net

Nowadays an AI agent is capable of fully exhausting the textual content written by a single person very fast and write as much code from it as is possible. Then it will start regressing to the mean with a lack of informative input and will start working towards something that has already been made many times. You will need to be ready to trawl much knowledge, get many domain experts in, to actually produce something meaningful. Normally this gathering of requirements has happened over a longer time scale, now it's compressed. Compression of the time scale means you'll need to arrange this to be super intense.

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

The software engineer work has changed with AI tools. AI engineers have already become used to the mode of operation where you need to keep the machines working on valuable things, otherwise they are only depreciating in value. Software engineers are becoming to be like that as well with AI assistants; they need to keep them working, and not only that, working in a way that creates value. This has always been what managers have been doing, keeping software engineers productive. It requires not only different skills, but also a different way of understanding work as it transforms from labor-intensive to capital-intensive. Have you noticed a change in how you understand value creation when you have a machine creating the value? #AI

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

How to Not Get Hacked Through File Uploads https://www.eliranturgeman.com/2026/03/14/uploads-attack-surface/

eliranturgeman.com

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

Israeli Software Used to Assassinate Ayatollah Khamenei Uncovered in Russian Surveillance Networks — UNITED24 Media https://united24media.com/latest-news/israeli-software-used-to-assassinate-ayatollah-khamenei-uncovered-in-russian-surveillance-networks-16854

united24media.com

Tero Keski-Valkama @tero@rukii.net

· 7mo ago

GitLab Threat Intelligence Team reveals North Korean tradecraft https://about.gitlab.com/blog/gitlab-threat-intelligence-reveals-north-korean-tradecraft/

about.gitlab.com

Tero Keski-Valkama @tero@rukii.net

· 4mo ago

Atmospheric Code Red: 2026 Super El Niño Now Trending Toward Record-Breaking Intensity » Severe Weather Europe https://www.severe-weather.eu/long-range-2/super-el-nino-2026-record-breaking-intensity-forecast-weather-impacts-united-states-canada-europe-fa/

severe-weather.eu

Tero Keski-Valkama @tero@rukii.net

· 5mo ago

Hantavirus outbreak: Cruise ship passengers had disembarked https://www.ctvnews.ca/world/article/dozens-got-off-cruise-ship-dealing-with-deadly-hantavirus-outbreak-after-first-death/

ctvnews.ca

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

Despite Doubts, Federal Cyber Experts Approved Microsoft Cloud Service — ProPublica https://www.propublica.org/article/microsoft-cloud-fedramp-cybersecurity-government

Federal Cyber Experts Thought Microsoft’s Cloud Was “a Pile of Shit.” They Approved It Anyway.

ProPublica

Federal Cyber Experts Thought Microsoft’s Cloud Was “a Pile of Shit.” They Approved It Anyway.

A federal program created to protect the government against cyber threats authorized a sprawling Microsoft cloud product, despite the company’s inability to fully explain how it protects sensitive data.

Tero Keski-Valkama @tero@rukii.net

· 6mo ago

With agentic engineering it is easy to just keep adding features, which leads to an incoherent and unsustainable whole even with perfect agentic engineering models.

Product managements is becoming ever more important, and someone needs to be able to look at the stakeholder needs and value propositions, and prioritize and plan on what is actually needed and what brings more trouble than value. This can be partially automated, but someone still needs to understand where the value comes from.

"Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away."
- Antoine de Saint-Exupéry

If you understand this and have experience with medium to large projects developed with agentic engineering practices, let's talk! I am hiring top agentic AI developers living in #Spain.

#AI #hiring #FediHire