Andreas Lundgren

There is this joke in software development: the first 90% of the work takes 90% of the time, and the last 10% takes another 90%. We humans tend to think linearly about processes (just look at how we assume tech trends will just keep on going), but the joke does reveal a basic truth about complex systems. Sweden's police force, despite doubling its budget and staff, finds itself paralysed by an increasing number of internal controls. The Swedish police now have at least 589 separate regulatory documents, many contradicting each other. As one officer describes it, the force has become "an elephant that only eats Excel sheets." I see this transformation of sensible oversight into bureaucratic quicksand everywhere, and it is not surprising that The Swedish Work Environment Authority has recently started to look into digital work environment and found it severely lacking. Getting from 90% to 100% compliance costs more than getting from zero to 90%. Administrators and policymakers, perhaps dazzled by visions of perfect control and often with a lack of understanding of how the work is actually done, keep building centralised systems that ignore this basic fact. The Police chief's admission that she doesn't even know the number of her own regulatory documents is a pretty good implication of how deep the dysfunction is — a system so complex it defies its own accounting.. In accordance with Lessig's rule "Code is Law", digitalised documents usually demand perfection, leaving individuals to translate a messy reality to perfect checkboxes that have been designed by someone who does not pay the cost of filling them in. Furthermore, when you demand that users fill in a form that where many cannot adequately describe the reality they see, or where the question is uncertain, you will also fail to get working decision support - you don't need a lot of bad data to contaminate the entire data set. Humans can ask questions and handle complex answers, forms usually can not. Modern bureaucracies, quite intoxicated by dreams of perfect oversight, have engineered their own paralysis. Their digital systems demand perfect data where human judgement was once enough (a parallel here is how we no longer trust experts) Smart governance means knowing when good enough is, well, good enough. The emergency measures now allowing Swedish police chiefs to ignore governance documents highlight the absurdity of all of this. In an age of constrained resources, we can no longer afford to let perfect be the enemy of good, especially if we want to meet recent challenges and disruptions. The alternative is clear: doubled resources yielding diminished results, as public servants drown in paperwork instead of serving the public. The Police force may be the ones called out, but really, this is everywhere.

35

5 kommentarer

Schibsted is splitting into Media and Marketplaces. My team, in Media, is shuffling between Developer Foundations, and Strategy & Enablement. My title is "tech enabler", but a title is just a name that we give something. It's not super clear to others what that involves. It's very much a role where I adjust based on what's needed at that point in time. Most weeks involve quite a wide range of different initiatives. Here's some of them: 1) A central service was shut down. The org needed to understand what led to the shutdown, and why. So, we talked to everyone involved, and produced a report that's 1/3rd of the length of the first Harry Potter book, summarized in a way that anyone could come in and find the relevant parts. 👨‍🏫 If we do not understand the organisation, and why initiatives fail or succeed, we will just repeat it over again. We've seen this document has been actively read, and we've given talks on it in several contexts. But talking to people, and understanding the human level of the org, is as important as writing about it. 2) Help connect tech leads, architects, and devs, through structured forums. 🚂 Schibsted Media is an organisation with many different brands and products. A lot of the work they do has overlaps, so ensuring they have contact points can help share ideas and drive collaboration. 3) Identify what tech teams are looking into, and what they should be looking into. Summaries of relevant tech news for us, with an explanation on why it's relevant. 📶 If the organisation doesn't know what libraries, frameworks, or languages teams are using, then decisions might not make logical sense compared to what's actually happening. Spotting potential collaboration and training is much easier with insight. 4) Spar, discuss, debate with anyone and everyone who wants to, and some that don't. Facilitate discussions, or workshops. 🎨 Whenever someone wants feedback on an idea, I'm there. I likewise do the same: I want to be challenged, to hear why someone might disagree with me! 5) Document practices, and write decisions. ⚽ Shared practices that aren't documented end up being quite wildly different, or forgotten about. Documention provides consistency, with insight into changing how they are done. 6) Identify cultural changes that might be a good idea, and try to influence things in that direction. Shine light on the good things happening. 🏴󠁧󠁢󠁷󠁬󠁳󠁿 I've worked in Media a while, so I have a decent idea of what our problems and successes are. Since I "float", I can influence a wide range of teams. 7) Fill the gaps in central tooling: mailing lists, channels, indexing of documents. 📜 I don't often write code for work now, but when I do, it's central. 8) Social committees and event planning 🎉 Having fun at work is one of the greatest "enablers" that could be put into practice. ... and a lot more. I write regularly about it here: https://lnkd.in/ddmNnTqx

21

1 kommentar

𝐂𝐨𝐧𝐬𝐮𝐥𝐭𝐚𝐧𝐜𝐢𝐞𝐬 𝐯𝐬 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐜𝐨𝐦𝐩𝐚𝐧𝐢𝐞𝐬 I've worked in 𝐬𝐭𝐚𝐫𝐭-𝐮𝐩𝐬, IT 𝐜𝐨𝐧𝐬𝐮𝐥𝐭𝐚𝐧𝐜𝐢𝐞𝐬, from niche to mid-sized to Large, the IT dept of non-software companies, and software 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 companies. Let me share some 𝐩𝐫𝐨𝐬 and 𝐜𝐨𝐧𝐬 about working in the different ones. 𝐂𝐎𝐍𝐒𝐔𝐋𝐓𝐀𝐍𝐂𝐈𝐄𝐒 I'll be honest: consultancies have a strong hold over the tech sector in Norway, as there simply isn't many or any big software product companies to speak of. Only small/start-ups. So, a lot of the talent pool is working in consultancies. I eventually left my home country to work in a Software 𝘱𝘳𝘰𝘥𝘶𝘤𝘵 company 🏡 💼 🛫 🌍 😁 𝐏𝐫𝐨𝐬: ➕ Work with experts & technology specialists ➕ Access to great training programs, technical and non-technical ➕ Exposure to many different industries! - Public and private sector. Retail, airports, energy, agriculture, government - etc! 𝐂𝐨𝐧𝐬: ➖ Tech > Problem-solving: You're sold as a <insert tech> dev. ➖ Process & Red tape - Pre-defined deliveries/contracts means things can be slow or difficult to change ➖ Little to no ownership of the product ➖ Certifications > Impact - Consultancies wants you to be easy to sell. Certs works as a proof of your skills. 𝐏𝐑𝐎𝐃𝐔𝐂𝐓 𝐒𝐭𝐚𝐫𝐭-𝐮𝐩𝐬 I had an internship in a start-up as a CS student. This experience was invaluable. I later worked as a software engineer in a start-up, and worked on some very cool projects. Had a lot of fun, learned a lot. 𝐏𝐫𝐨𝐬: ➕ Fast paced, not boring ➕ You can and must wear many hats ➕ Could be career making 𝐂𝐨𝐧𝐬: ➖ Little to no structure or process ➖ Potentially nowhere to grow ➖ A big risk that might not pay off. If you're in a start-up, make sure to get something out of it even if the start-up doesn't succeed. 𝐍𝐨𝐧-𝐬𝐨𝐟𝐭𝐰𝐚𝐫𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐜𝐨𝐦𝐩𝐚𝐧𝐢𝐞𝐬 I worked in television(!) As a Business Intelligence dev & analyst. Basically a data engineers & Data scientist. ➕ Can have big impact on the product/business ➕ Exposure to potentially interesting domain ➕ Help non-technical stakeholders ➖ Not solving for the core business, not prioritized ➖ Slow to adopt new tech. ➖ Lonely - No dev community. Things like Code quality, CI/CD, DevOps, DevEx... is up to you/your dept, or nonexistent. 𝐒𝐨𝐟𝐭𝐰𝐚𝐫𝐞 𝐩𝐫𝐨𝐝𝐮𝐜𝐭 𝐜𝐨𝐦𝐩𝐚𝐧𝐢𝐞𝐬 I might be biased here as I currently work in a Software product company, something I wanted for a long time. ➕ Have a big impact on the core business and or for customers ➕ Win by solving interesting & challenging industry problems in smart ways over use <insert tech> by all means. ➕ Ownership of product over time ➕ Career growth opportunities ➖ Working in a high performance environment can be stressful ➖ Bias for action or bust ➖ Learn how to deal with Ambiguity or have a bad time Please let me know what you think of this!!!

31

9 kommentarer

This week I spoke to Klarna CEO Sebastian Siemiatkowski. Sebastian led Klarna from a small profitable business to a $50B valuation. In our discussion, Sebastian shares how he turned Klarna back to profitability and discusses Klarna’s ongoing pursuit of creating a digital financial assistant. He also reflects on lessons from the "Succession" boardroom drama and outlines the operating principles that have made Klarna the company it is today. → How Sebastian boosted productivity at Klarna In the early days, the three Klarna co-founders faced challenges over perceived differences in work effort. So, they repositioned their computers so they could see each others' screens and started tracking their hours. The policies discouraged distractions and helped the founding team realize they could spend a bit more time doing productive things. Just like a personal trainer overlooks sets to help motivate you, accountability over working time helped Klarna boost productivity. → Don’t be too service-minded Sebastian notes that some of the best leaders are driven by a desire to serve others, a trait rooted in empathy. However, this service-mindedness can lead to conflicts and burnout, as leaders might struggle to say no and try to fix every issue raised around them. However, sometimes saying no is necessary and not all problems need immediate fixing. This balance between empathy and assertiveness is crucial for sustainable leadership. → Are small teams better? In pursuing an ambitious $100B+ opportunity of creating a digital financial assistant, Klarna realized the need for improved operating execution. They shifted from a traditional functional setup to autonomous, small "startups" within the company, influenced by Amazon's two-pizza team concept. Each team, consisting of eight people, was tasked with writing a press release detailing their future accomplishments. While this approach initially boosted productivity and agility, it also led to a lack of coherence and standardization, with teams working in different directions. Now, Sebastian acknowledges the value of agile methodologies but also recognizes they need to be balanced with consistency and standardization within the organization. → Valuation Volatility To aggressively pursue winning the U.S. market, Klarna raised money fast and started burning cash rapidly. High-profile investors quickly boosted their valuation from $5 billion to $10 billion and soon to $30 billion just a year later during COVID-19. However, the rapid valuation growth outpaced their actual business performance, and Sebastian acknowledges getting caught up in the exuberance of the market. Despite the challenges of a significant down round, their heavy investment in the U.S. market paid off, as it became their largest and profitable market, and the rollercoaster taught Klarna to become more thoughtful with resource allocation. More insights in the full episode: https://lnkd.in/dfvupFGA

84

1 kommentar

Both Ilya Sutskever and Jan Leike left OpenAI yesterday, with Ilya announcing that he will soon be revealing a new project. OpenAI estimated that Superalignment will be solved within four years. This means either they have successfully solved it, or else have realised that they are on the wrong path, and OpenAI won't be able to solve since focus has now shifted towards product and distribution. As for what superalignment or alignment means in general? In layman terms, whenever you enter some prompt in ChatGPT, you expect a particular answer in return. But often ChatGPT misunderstands and gives you the wrong answer. In some cases it might give you an approach that you didn't ask or is different from what you expected. Ex: Asking ChatGPT to calculate the area of a triangle from the given dimensions using Heron's formula. ChatGPT finds that the triangle is equilateral, and instead uses √3/4*a². Therefore the goal of superalignment is to ensure that Artificial Intelligence models stay within the bounds of human value, and their goal aligns with that of the requirements of a human, thus avoiding any unnecessary harm. Basically risk mitigation.

9

Have you tried MVW pattern in your apps? Model - View - Whatever? Jonas Uekötter was recently featured on Sean Allen's Swift news for the month of April. In his twitter thread he points out that when choosing a pattern, the real issues often stem from overcomplicating things. This is a refreshing reminder for us all: - Keep it simple - Understand your goals - Use tools that suit your team Have you ever felt overwhelmed by the many ways to structure app development? What's your approach to choosing development frameworks and patterns? #MobileApp #Architecture #iOS #Swift #Apple #Google #MobileAppDevelopment #Flutter

108

39 kommentarer

Flow: Startup wants to multiply every CPU performance with an additional chip A Finnish startup is currently coming up with a statement that is hard to believe: They want to be able to increase the performance of a processor many times over - even up to a hundred times. All they have to do is integrate a newly developed additional chip. Theoretically known for longer The company Flow is a spin-off of VTT, a research facility supported by the Finnish government. The startup's employees have developed a parallel processing unit (PPU) that is intended to give a CPU a helping hand. According to the developers, it doesn't matter what basic design the processor is based on, so the PPU should be able to accelerate x86, ARM and other chips. Of course, you can't just get as much performance out of a CPU as you want. If it were that easy, the big manufacturers would certainly have done it already. Nevertheless, computer science basically knows ways to organize the work of a main processor more efficiently. The problem is that CPUs - unlike GPUs, which are optimized for parallel calculations - still work serially. Since the days of tube computers, nothing has changed in terms of the fact that they can only carry out one operation at a time - even if they have of course become considerably faster. "The CPU is the weakest link in data processing," said Timo Valtonen, co-founder and head of Flow. "It is not up to the task, and that must change." The PPU Flows cannot change the serial work either. However, the additional chip ensures that the consecutive operations are made available for processing in an optimal order. It also relieves the CPU of various trivial tasks so that they do not slow down the computing time. Integrated into the chip The principle of this has been discussed in computer science for a long time. However, practical implementation has so far failed mainly because the known approaches ensured that only newly written, specially optimized code could be used. A chip system for which all software would have to be rewritten would certainly have no chance on the market, even with higher performance. Flow's great achievement is not just the management of high-speed data traffic, but the fact that this is possible without changing the code on every CPU or architecture tested. However, the PPU must be tightly integrated into the processor's chip architecture - it is therefore not possible to simply plug it into an expansion slot on the PC or server like a coprocessor. If Flow's benchmarks are correct, there are sure to be chip manufacturers who will take on this effort. Flow says it has seen up to 100-fold increases in code that has been modified (though not necessarily completely rewritten) to take advantage of the technology. The PPU worked as part of an FPGA under ideal conditions. Much lower increases are usually achieved - but even doubling performance would be a huge step.

1

There are many ways to benchmark LLMs - is there a consensus which benchmarks can be used for Nordic languages or should we create one? The current ones used in research often rely on logprobs so can't be used with API based models. PS. phi-3 performance is quite wild

30

10 kommentarer

Klarna announced that they're filing for a U.S. IPO - https://lnkd.in/dd_dUpBN As Fintech professionals, why should we be paying attention (especially those of us in startups and growth-stage companies): 1. Market Validation, Investor Confidence, and Valuation - Klarna’s move to go public underscores the growing maturity and acceptance of fintech solutions. A successful IPO could boost investor confidence, potentially leading to increased funding and opportunities for other fintech startups. Lastly, Klarna’s IPO valuation will be closely watched as an indicator of market sentiment toward fintech companies. A strong performance could lead to higher valuations for other fintech firms, while a lackluster debut might prompt more cautious investment approaches. 2. Regulatory Scrutiny and Compliance - As a publicly traded entity, Klarna will face heightened regulatory oversight. This may prompt stricter compliance standards across the fintech industry, encouraging companies to invest more in regulatory frameworks and transparent operations. 3. Innovation - With the introduction of shareholders and growth expectations, Klarna might diversify its offerings beyond traditional BNPL services. This could set a precedent, inspiring other fintech firms to innovate and expand their service portfolios to attract and retain customers. #FinTechBrainTrust #Fintech #Payments #Startups

11

Such a great explanation why your AI generated talk abstract gets rejected. This one is about NDC Oslo, but I can assure that we at AWS Community Day DACH think very similarly about abstracts. I absolutely like that quote from Mathias Verraes: > Writing is thinking, and if they outsource their thinking to a machine, I'm not interested in what they have to say Thanks Einar Høst for that talk and ☁️ Anders Bjørnestad for sharing with me. Watch that video: https://lnkd.in/dpuDXbe4

30

3 kommentarer

At Slush last week, I heard some investors received over 1,000 meeting requests. It’s a staggering number 😶 I couldn’t help but feel amazed. When I sat down with VCs focused on AI, their amazement matched mine. They were impressed that we’ve built our own AI with our own team, rather than relying solely on tools like OpenAI or Google APIs. In a space where 90% of AI startups use pre-existing platforms, this distinction stood out. There’s nothing wrong with applying AI to create value—many startups do it brilliantly—but it’s clear we’re talking about two very different skill sets: - Applying AI: Innovating through existing tools. - Building AI: Creating the models and systems that push boundaries. Both approaches are vital, but they deserve separate recognition. As the AI field grows, perhaps it’s time to celebrate these unique contributions in their own compartments. For founders, the real magic is in creating something remarkable, not in chasing popularity. 💫 #AI #Startups #Slush

48

1 kommentar

After much hard work, today we officially released our user facing EU-election AI chatbot "Valkompisen": https://lnkd.in/d-WbXn63 . This project is a cooperation between Aftonbladet Editorial Team and the Aftonbladet AI-hub team. The "Election Buddy" is based on OpenAI's GPT-4 technology. To ensure the provision of relevant and accurate information about the EU and the EU elections, we have created a database containing facts reviewed by Aftonbladet's journalists. It also includes responses to multiple surveys answered by the eight parliamentary parties and their top 10 candidates. A significant part of the database also consists of facts about the EU election, the structure of the EU, the history of the EU, current topics in the EU debate, votes in the European Parliament, party groups, and much more. When a user asks a question, the AI bot performs a semantic search in the database to retrieve relevant facts. The bot then combines this information and formulates a well-founded response. The Election Buddy has also been provided with guidelines, or "prompts," that direct how it should use the information to respond and what tone and personality it should convey. To ask questions you can press the predefined questions or login to ask your own questions.

42

4 kommentarer

Did you know that the same mid-level software developer hired full time in Stockholm costs the employer about €49.7/hour versus €52.5/hour outsourced to us? (More on the TCO calculation and assumptions in our blog post) "Outsourcing is expensive" is a common argument when pitching to potential new clients and one that deserves attention. Once you take the price out of the equation you can truly focus on what matters when crafting your SW development team. Yes, you want a reliable team crafting high quality code. But more importantly, you deserve a team that will help you with smart decisions to not under-engineer or over-engineer individual parts of the solution and that comes with vast experience and the collective knowledge of the entire company you get every time you work with us. Have you heard of the price-quality-speed triangle? Yes, it's possible to strike a balance of all three and get access to meticulous skill set, flexibility, speed, and even potential cost savings that comes with these. Just give us a call. https://lnkd.in/gA7FTsVz #software #craftsmanship #roi #smartdecisions #indigolabs

15

Presenting and demoing the early stages of Popcorn Cloud yesterday was awesome. I'm thrilled to have Anton Ganhammar and Axel Eriksson as part of the team. What we're doing is _not_ your average no-code thing—and yesterday we made sure to express that. We are truly in the business of "developerless", which like serverless, is not about "not having developers" but being straight about the fact that we shouldn't have to build IT _around_ developers (soon): That's our very clear mission. Make sure to DM me or the team if you want to learn more. This _is_ the next big thing from Gothenburg. #saas #startup #locode #nocode #developerless #cloud

103

12 kommentarer

You may have seen the recent news regarding OpenAI releasing a voice for ChatGPT that sounded like Scarlett Johansson – which is a little awkward given that they approached her, and she declined. In better news, it inspired me to re-watch Spike Jonze's prescient 2013 movie Her. I can safely say it's quite a different experience today than it was then. The technological advances have turned much of science fiction into reality, making the questions it asks about our humanity even more relevant. 1. How can technology bring us together, and how can it come between us? 2. How do we as linear humans navigate exponential technological change? 3. And above all, where can I get Joaquin Phoenix's outfits from? We can make a better future happen – so let's choose to. With empathy, thought, and design intent, we can harness AI to elevate the human experience. Watch the trailer: https://lnkd.in/gktNJceg #AI #HerMovie #ScarJo #OpenAI #ChatGPT #technology #exponentialchange #design #artificialintelligence

32

10 kommentarer

I love Backstage from Spotify But I feel that people are given a fantastic tool, but not guidance on how to use it. We’ve had platform engineering for a couple of years now, and now the new hype is Developer Experience. This is where Backstage is the go-to platform. But, we keep on making the same mistakes again and again in our industry. Do you have a strategy for your backstage offering ? Who is your target audience? What do they NEED ? Do they need a Flux plugin ? Do they need a Kubernetes plugin ? NO. They need to create products, ensure they are ok and release new features. It needs to be developer centric. You need to understand their domain, needs, problems and concerns and then think about how you can ease or provide help with Backstage. Make the responsibility clear on both the developer and platform side. Who is responsible for what ? What signals do they need to react to the things they are responsible for ? Anything more then that is noise. Confusion. Build a Backstage platform your developers need, not the one you need.

100

6 kommentarer

-Didn't you get the memo? - eeeeh? What memo? - The CoDeMeMo - Continuous Delivery Metric Model I'm excited and happy to re-publish the Continuous Delivery Model. This model originates from the time when I founded (2007) and operated Praqma. A company specialized in Continuous Delivery implementation of software development processes. Praqma is now gone. The company was swallowed in a merger and acquisition process in 2019 and the model didn’t survive. It was stored away in a GitHub repository and never used again. ...until now! The model was optimized specifically to be used as an assessment tool for helping companies to map their current Value Streams and to build a backlog for implementing Continuous Delivery. The assessment process was based on “The Toyota Way” principles: #ValueStreamMapping, #GenchiGenbutsu (observe deeply), identifying waste (#muda, #mura and #muri), measure #LeadTimes and Continuous Improvement (#kaizen) based on #5S (sort, straighten, shine, standardize, sustain). Lean Software Development! The assessment process we developed would span five intense workdays on-site, with a mix of all-hands-on-deck Value Stream Mappings, interviews, observations, and lean coffee workshops. Some of the larger assessments we did had 120 participants in the first-day Value Stream Mapping workshop. I’m now re-publishing the model. I had to rewrite the code as I didn't have acces to the original codebase. I enhanced readability and responsiveness, but more importantly I’ve turned the model into an open source template you can inherit from in #Jekyll and host it as a static web. Build your own model! ...More on that later! But for now I just want the release this MVP so we can get the model live again. How To Read The Model: Its a matrix with six different areas to consider in becoming black-belt Continuous Delivery - Software as a Factory champion. Each area has 5 levels of expertise ranging from novice to expert. Each card in the matrix represents a certain theme or disciplin. And each card is scored - by the team itself, during the workshops - for effect on three parameters: - Throughput (lead times) - Feedback (data for decision making) and - Payback (...on technical debt) And then scored for complexity as one of: - A low hanging fruit - A team effort - A mountain to climb To show-case the features of the model I have added an anonymized and generalized write-up of some of the typical or illustrative findings and mitigation from the total of 24 assessments that I conducted on the basis of this specific model. I've started out with the three first cards in the "Build" area - more will follow. You can subscribe to the atom feed in the footer on the website to get updates. I would love your feedback on the usefulness and suggestions for improvement. If you and you team would like to do this assessment on your product and process, I'm happy to announce that I'm back in business, charged and ready - just reach out!

39

18 kommentarer

Next week (Thursday 26/9) I'll once again be haunting the Göteborg React Meetup with silly ideas and dark thoughts—more specifically, I'll be there presenting "Entropy, Rot, and Decay in IT and Software". You'll learn how entropy is a quite tangible, handy concept to help understand why organizations deteriorate and why technology orgs specifically rot, and eventually, decay into obsolescence. Like in the "whisper game", meaning breaks down when there are too many connections—big organizations struggle to be relevant, fast, and innovative: But they could be! If you ever thought the universe is against you, I will explain that "yes, it is". But that's OK, as you will also learn.

27

4 kommentarer

Ilya Sutskever has just launched a new AI venture called Safe Superintelligence (SSI), aimed at developing safe and powerful superintelligence. This comes one month after he left OpenAI. SSI is co-founded by Sutskever, former Y Combinator partner Daniel Gross, and ex-OpenAI engineer Daniel Levy. The sole focus of SSI is to create beneficial superintelligent AI, which Sutskever believes could emerge within a decade. SSI will be a pure research organization, insulated from short-term business pressures. #ai

2