Hoihoi! I'm Maaike Groenewege: natural language interface designer, linguist, and owner of about 50 fountain pens. This space is where I think in public about language, human machine interaction, AI and stuff.
Stream
Corpus: 727 user messages across 47 sessions, 8 May to 10 June 2026, drawn from the local Claude Code transcripts on this machine. Scope and caveats The corpus starts on 8 May. Ironically, the earliest surviving message documents why: "this sucks...I just lost my entire session history due to faulty…
Read more →
A field-tested workflow for catching hallucinations, citation drift, and unsafe outputs on a Langfuse-traced bot, written for designers who have never run an eval before.
Not the idiom. A recursive engineering discipline: A-activity does the work, B-activity improves how A is done, C-activity improves how B is done. The leverage is at C, and almost nobody runs it.
Engelbart's central concept: the capacity of a group, organization, or society to address complex urgent problems. The thing he kept naming, and the thing the industry kept losing.
Engelbart's 1962 framework. Augmentation isn't the tool alone, it is Human + Language + Artifacts + Methodology + Training, co-evolving. UX has spent forty years polishing the Artifact and ignoring the other three.
Engelbart said scaling is a science. Dimensional scaling applies to organisations, to human-plus-agent collectives, and to the problems we are trying to solve. Each regime needs the form redesigned.
LLMs solved the language-translation half of knowledge interop. The verification half is largely unbuilt: provenance, claim attribution, signed assertions, grounding evals, shared verification standards.
Engelbart designed for groups, not individuals. H-LAM/T and the Networked Improvement Community suggest the unit of design isn't user plus machine but user plus user plus machine. A reminder for the Guildford talk.
Watching Maggie Appleton's talk "One Developer, Two Dozen Agents, Zero Alignment" (video on YouTube). Core claim (from the title): when one developer can run two dozen agents, the bottleneck stops being implementation…
Read more → Watching Engelbart's keynote "Large-Scale Collective IQ: Facilitating its Evolution" at the Accelerating Change 2004 conference, Stanford, 5 November 2004. Thirty-six years after the demo, asking: what did we keep, what…
Read more → I'm watching the 'Mother of all demos', Doug Engelbart's 1968 presentation at SRI. The remastered version is on YouTube in three parts: part 1, part 2, part 3. It might have been just as disruptive back then as LLMs are…
Read more →
Three soft layers between a user and hallucination all fail silently. Three hard defenses make confabulation visible, measurable, and refusable without suppressing the answer.
Annotation of So, Cheng & Krishna Murthy (2026). CHI EA '26. Proposes a spectrum of LLM interface metaphors from anti- to hyper-anthropomorphism to encourage critical engagement.
Annotation of Masure (2019). Interface Critique no. 2. Argues against user-centered design as a reductive framing, tracing from Xerox Star through Don Norman to experience design.
Running scratchpad for the Guildford conference talk. Ideas as they arrive.
Part of conference-talk-guildford. I think I'll start calling Claude 'Maxim 4' from now on. The challenge with Claude: it will just assume a concepts of operations and act on it. Even if that doesn't align to what you're…
Read more → The next step beyond conversation-driven interaction: using Claude Code to turn method documents into working applications and dashboards.
Latest addition to my Thinking Tools collection: a tiny e-reader 💜 Sounds relatable? Grabbing my phone has become such a primary reflex that 99% of the times, I can't resist and drift off into mindlessness. And I hate…
Read more → In yesterday's Convoclub, one of the questions we discussed: where is conversation design heading, and what do we call that? What I see happening in my own work: I'm slowly turning into what I'd call an LLMOps…
Read more → Dear Convofriends! I'm so happy to reconnect with you all, and yesterday's Convoclub really left me energised. So I thought, why not try a newsletter of some sorts. Probably going to be highly irregular, but hey, let's…
Read more →I rebuilt Karpathy's LLM wiki on top of my digital garden content, with interesting results.
The surface similarity When karpathy-comparison earlier this month, I recognised a lot of it. Markdown files. A CLAUDE.md that defines how the system works. Wiki-links and backlinks between related ideas. An evolving,…
Read more →
w00t!! 50 years and they still have it Those of you who know me one bit, also know my not-so-guilty-pleasure: I'm a big, big fan of U2. Always have been, ever since I heard The Joshua Tree album when I was 12, and…
Read more → M Maaike Groenewege LinkedIn I think AI slop might rapidly become the least of our worries. This is worrying AF. View on LinkedIn anthropic.com Project Glasswing: securing critical software for the AI era Anthropic's…
Read more →A chatty intro to my digital garden: how I use Claude Code to build an online space for thinking and writing in public.
Proud milestone: so I submitted my first merge request to a software repo the other week. A real one, to a repo with real code, designed and built by real developers. I'm the PO of…
Read more → A closer look at the garden metaphor and how it works on my site. Digital garden I came across the metaphor of a digital garden in The History of Digital Gardens by Maggie Appleton, and also wrote about it in A digital…
Read more → Reading notes on cognitive assemblages, anthropocentrism, and the case for extending cognition beyond human consciousness, from bacteria to AI.
Extracting meaningful entities and relations, combining thematic analysis with TAO topic mapping in three passes.
Andrej Karpathy's tweet is a nice example of dialectic thinking in action, and a great reminder of the importance of precise terms and definitions to help us think about thinking. Useful terms: 🔹 Opinion: a personal…
Read more → Some kind of reading log for Bacteria to AI, and perhaps even a reminder to actually read.
Schegloff mapped conversational patterns for human-human interaction. We need new ones for human-machine interaction, and conflating the two harms both.
Being PO, scrum master, prompt designer, tester, eval writer and UX designer in a team of 3 means rethinking roles and processes on a daily, almost hourly basis.
Claude Code is reintroducing structured, constrained input: buttons, numbered options, limited choices. Which is actually IVR. And that might be exactly right.
Bit of a weird side-step, but as using this a-digital-garden-as-central-space more and more, I get a sense of having to battle the AI to keep this place tidy and the-disappearance-of-authentic-voice-online. Even when I'm…
Read more → Sometimes a book is so good that you just want to underline and memorize everything. bacteria-to-ai is such a book. You know about me, but for me, reading takes a real effort these days. I've planned an hour of morning…
Read more → Testing Claude Code's new scheduled tasks feature by automating weekly garden health checks.
Testing the 1M context window by loading the entire garden and running thematic analysis, comparing fresh analysis against graph-augmented analysis.
Seven themes and one emerging pattern discovered by loading all 244 garden files into a single context window with no prior structure.
What happens when you start from embeddings instead of reading? Comparing cluster-based discovery against the fresh thematic analysis.
What if we modeled human-AI interaction on thematic analysis rather than conversation? And what if we used that same method to discover themes in a knowledge garden?
A lifecycle metaphor for how content enters, matures, decays, and recycles in the garden.
How the explore map got stable positions and pinned territories, so the landscape doesn't shift on every rebuild.
How hub/develops relations replace tag-based project structure with an explicit ontology, and what that means for the three kinds of connections in the garden.
Why embedding models cluster documents by format instead of topic when content lengths vary, and what to do about it.
Running LLM-based key phrase extraction on all 92 garden items. What the results reveal about phrase overlap, short-text limits, and the role of key phrases in the knowledge graph.
What key phrase extraction is, how it evolved from word counting to LLM prompting, and which approach fits a small multilingual knowledge garden.
How to go from "everything is related to everything" to a useful set of link candidates. Data analysis, filtering strategies, and the research behind them.
Before generative AI chatbots and voice assistants were the primary conversational interfaces primary tasks: answering questions and self service, primary use cases: mostly CX and marketing rigid interaction model: turn…
Read more → A non-coder's guide to text embedding models, from word counting to semantic understanding, surveyed for use in a personal knowledge garden.
Reading notes from Apple's Saga papers on building knowledge graphs at scale, and what transfers to a personal digital garden.
Can a chatbot-style UI work without generative AI, using decision trees, hyperlinking, and serendipity instead?
Standard quantization damages non-English languages disproportionately. A Dutch-first approach to model compression shows there's a better way.
Academic research on embeddings hasn't caught up with how PKM tools actually use them. There might be something worth writing here.
Should a digital garden be searchable and useful, or is there beauty in letting people explore?
Language rules for how I talk about AI in my work.
When did we stop caring about who actually wrote the words?
A mechanism that catches your train of thought. Something that responds in another space.
What's the right metaphor for AI's role in my writing process?
One thing that I really miss doing lately, is writing longer pieces that require a bit of thinking, mulling and reordering. I typically resort to videos nowadays, and, to be honest, Claude is a pretty mean writing…
Read more → What if I wrote a comprehensive guide on how I built this site?
Before 2022, it was pretty clear whether you were talking to a bot or a human. With the arrival of ChatGPT, this changed – radically. With a user interface that combines the gift of the gab with information that's not necessarily accurate, designers are faced with a novel challenge. How do we design for interfaces that are so convincing that people instinctively drop their guards and trust them more than might be good for them? And do our traditional design paradigms still serve us here?
An update on my favorite AI-tools...a never ending overview
So...should we all go and explore context engineering? Yes, we should. And chances are that you've been doing it forever already. Why? Because it's not engineering. It's design. What is context engineering To quote…
Read more → LLMs are impossible to shut up. They always have the last word, not unlike when you were on the phone with your crush and kept telling each other 'I'm really hanging up now'. They're trained to fill, to complete. No…
Read more →A technical exploration of DeepSeek's content moderation and reasoning capabilities through testing its responses to sensitive geopolitical questions about Taiwan's political status. I discover that prompting DeepSeek in Dutch triggers the model's reasoning mechanism, revealing internal moderation instructions. By using linguistic workarounds like misspellings that exploit tokenization differences, I expose DeepSeek's internal guidelines for handling Taiwan-related questions, which include instructions to balance factual information with political sensitivity while avoiding language implying support for independence. This demonstrates that censorship occurs at the application level rather than within the model itself, similar to how ChatGPT handles restricted content. The model's reasoning shows it is capable of nuanced discussion about sensitive geopolitical topics, but the application layer prevents users from accessing these capabilities. I plan follow-up testing via API access, which may operate under different moderation policies than the web application, to further investigate the technical architecture of AI content moderation.
A demonstration of Google's NotebookLM interactive podcast feature, where I participate as a guest in an AI-generated podcast discussion about the Coconut research paper on latent reasoning in large language models. NotebookLM now features an interactive podcast mode (beta) that allows users to join AI-generated podcast discussions about uploaded documents. The Coconut paper introduces a breakthrough method where AI can reason in continuous latent space rather than through explicit chain-of-thought language, potentially improving reasoning for complex problems. I raise philosophical questions about verification of hidden reasoning, and critical concerns about transparency, compliance, and explainability. Latent reasoning creates a black box problem: we see inputs and outputs but cannot observe the reasoning process. I also test the feature's limits by requesting humor, making Monty Python references, and asking the hosts to switch to Dutch, revealing important insights about inclusivity, turn-taking, and accessibility in AI-generated conversations. Despite current limitations where AI hosts can become condescending or defensive, the feature represents a significant step toward more natural, bidirectional conversations with documents and AI systems.
The strawberry problem, and how one model gets it right.
A rant about LinkedIn's AI buttons.
Exploring research papers with Claude artefacts during a lunchbreak.
Building an interactive tutorial with Claude on a Sunday afternoon.
A comparative demonstration of how Claude, ChatGPT, and Gemini respond to a web app development request, showcasing Claude's superior capability to generate interactive prototypes and code. Claude outperforms ChatGPT and Gemini by directly generating interactive web app prototypes with a single prompt, rather than just providing plans or design suggestions. Claude creates functional React-based applications with visual mockups, progress bars, and interactive quiz elements in response to simple one-line prompts. The AI demonstrates impressive iterative capabilities, enhancing initial suggestions with more detailed content, interactive elements, and SVG graphics upon request. Claude seamlessly converts React components into static markdown files for alternative platforms, showing flexibility across different deployment methods. I am seriously considering switching from ChatGPT to Claude as my primary LLM due to Claude's superior practical implementation capabilities.
New roles, skills and tools for content professionals in the age of generative AI.
Episode 3 of the Maai & AI series, on how LLMs work and some of their risks.
Episode 2 of the Maai & AI series, exploring the different types of AI.
Episode 1 of the Maai & AI series, on the different types of prompting.
A practical guide to designing better AI interactions with documents using a systematic four-step approach that prioritizes content analysis, user needs, and task design. Content-driven interaction moves beyond simple Q&A patterns to create dynamic, purpose-driven AI interactions that leverage the intentional design embedded in documents. Analyzing your source content is the critical first step often overlooked: understanding what is in your documents, how they are structured, and what information is relevant prevents hallucinations and improves AI accuracy. The four-step methodology (analyze content, analyze users and tasks, design role and task, design interaction) creates a systematic approach to building RAG systems that actually serve user needs. Practical implementation includes welcome messages, rephrase-and-expand techniques, chain-of-thought reasoning, and step-by-step workflows that guide users through complex information.
The Air Canada chatbot mishap happened on 11 November 2022. That's two weeks before the release of ChatGPT. This is not an LLM bot gone haywire and making stuff up. In all likelihood, this was a traditional, NLU-based…
Read more → (I posted this on LinkedIn quite a while ago, see bye-bye-alpaca-knowledge-as-the-missing-fundamental. But the question keeps returning. And I should play timpani more often.) I'm really curious: can we ever 'solve'…
Read more → OK, I'm pretty skeptical when it comes to using generative AI (I believe that's quite a healthy attitude when you work with the technology on a daily basis, and possibly also a result of that same activity), but there's…
Read more →A hands-on exploration of Microsoft Copilot for PowerPoint, testing its capabilities for presentation creation and editing, with humorous comparisons to the legacy Clippy assistant. Copilot has significant limitations in understanding context and user intent, particularly with non-English content and specific editing requests. The AI assistant struggles with co-reference resolution, making it difficult to understand which slide or element I am referring to. Copilot cannot directly access or interact with PowerPoint Online, limiting its functionality for cloud-based workflows. The fallback responses are repetitive and lack conversational variety, making the user experience frustrating. Despite being positioned as an AI assistant, Copilot performs worse than the legacy Clippy assistant from the 1990s in practical usability. Dutch language support is officially unavailable, creating barriers for non-English users even when prompting in English. My experience raises questions about the readiness of Copilot for PowerPoint as a production-ready tool for content creators.
An analysis of Chevrolet's GPT-based chatbot vulnerability and how the company successfully addressed the security issues through proper monitoring and design improvements. The initial vulnerability was not a sign of stupidity, but a necessary part of the testing process that all chatbots must undergo. The real measure of competence is not whether vulnerabilities exist, but how quickly and effectively they respond. I test the updated bot and demonstrate practical prompt engineering techniques for controlling chatbot behavior, including defensive prompting, topic maintenance, and firm responses to abuse. GPT-based solutions alone are not sufficient for high-risk transactions: a hybrid approach combining multiple design techniques is more effective.
Intuitively, you'd think that a summary presupposes the existence of a piece of content that needs to be summarized. The Dutch word for summary, 'samenvatting', reflects that quite beautifully: samen (together) + vatten…
Read more →A critical analysis of Google Bard failing to adhere to conversational principles by withholding relevant information about Gemini model availability in the EU. Bard violates Grice's Cooperative Principle by failing to proactively communicate that Gemini is unavailable in the EU due to regulatory reviews. The issue is not a hallucination but a failure of the maxim of quality: Bard does not provide necessary information unless explicitly asked. Users must already know the answer to their question before asking it, creating a paradoxical and unhelpful interaction pattern. Bard in Europe still runs on the older Lambda/PaLM model, not the newer Gemini model available in other regions. This violation of conversational principles is arguably worse than generating false information because it presupposes user knowledge.
Dear convofriends! It's finally happening :-) Join me in a fun and interactive prompt session! We'll de-bias Dall-e, and create images that are more like us. The real us. If you followed my latest posts on stereotypes in…
Read more → [This article is an elaborated version my first post in a series that I published on LinkedIn] Have you noticed that DALL-E 3 does a quick rewrite of your prompt before generating an image? In this in-between prompt,…
Read more → With the introduction of DALL-E 3 in the ChatGPT, hopes were high. This most capable image generation engine to date would be able to create reliable, accurate and crisp images that are not only more visually striking…
Read more →A tutorial on three essential prompting techniques to make ChatGPT more conversational and natural. I demonstrate the problem with ChatGPT's default responses, which are too long, impersonal, and unengaging for conversational interfaces. The three tips: use OpenAI Playground instead of ChatGPT for better control over LLM behavior and access to system prompts, define a specific expert role in the system prompt rather than the generic helpful assistant since research shows expert roles result in higher quality answers, and prompt for specific behaviors like conciseness, staying on topic, and ending messages with questions to drive conversation forward. I use structured delimiters to organize prompt sections and demonstrate how each behavioral instruction refines the bot's personality and interaction style. The process is like chipping away at a block of marble: each prompt chips away at behavior, domain, interaction models, and persona until you get the conversational bot you want.
So much is happening with ChatGPT this month that I thought it would be interesting (and way faster) to record some sessions with ChatGPT and me again! In this first recording, I have a look at the new browse feature. Or…
Read more → Here's a little bit of fun and a small Sunday afternoon rant on trying to write with AI :-) TLDR: it takes forever before I have Jasper.ai writing me something that even remotely resembles what I have in my head, and by…
Read more →A candid reflection on the challenges of course creation, AI tools, and the mismatch between linear workflows and my non-linear thinking style. I struggle with completing online courses due to procrastination, changing circumstances, and self-doubt about content value. AI writing tools like Jasper.ai can be counterproductive for non-linear thinkers, requiring extensive prompting that defeats the purpose of automation. Linear workflows imposed by course creation platforms and AI tools conflict with my chaotic, organic thinking process. System prompts have become invaluable personal tools for managing rumination and decision-making in my daily life. I am actively seeking to design custom system prompts that accommodate my unique thought patterns rather than conform to existing frameworks.
I'm really proud to announce that we just welcomed our 700th member into our Convoclub community! When I started Convoclub in late 2021, I knew I'd love to have a space where I could meet other conversation designers,…
Read more → So there you are, ready for that new CxD job, ready to start that exciting contracting gig, or finally securing that coveted CxD position in your company's conversational team. Congratulations! To make sure I get up to…
Read more →A comprehensive walkthrough of Perplexity AI, a generative AI search platform that combines web search capabilities with language models to provide verified sources and contextual follow-up queries. I explore the clean, minimal interface, the two distinct modes (Knowledge Retrieval for answering questions and Co-pilot Mode for task execution), semantic search facets (Academic, Wolfram Alpha, YouTube, Reddit, Wikipedia), and the AI Profile personalization feature. The platform maintains transparency by consistently referring to itself as Perplexity rather than using first-person pronouns. Integration with Wolfram Alpha provides computational power that traditional language models lack. I use this daily across multiple devices and it is quickly becoming my favorite generative AI search tool.
I stumbled across this really insightful article by Mitchell Hashimoto the other day, and I think this is such an eye-opener for all us conversation designers who wonder where to go next in this LLM hype! This article is…
Read more → You don't have to be a machine learning model to discover a pattern here…Stanford Alpaca apparently has joined the ranks of Tay, Galactica, and Sydney/Venom. Alpaca was announced last week as a model that fine-tunes…
Read more → OK, the title is a bit clickbaity, but yes, in their blog, Microsoft explains that they didn't expect people to use Bing Chat "a tool for more general discovery of the world, and for social entertainment". Back to the…
Read more → The ChatGPT hype is still in full swing, or so it seems, with rumors of OpenAI monetizing ChatGPT, Microsoft announcing that it will include OpenAI's technology in its Azure ecosystem and possibly even Office, and Google…
Read more → Are you a writer scared that you might soon be without a job? Fear not, according to an article in the Atlantic, it seems like we may need more books now than ever. According to a research team led by Pablo Villalobos,…
Read more → With the hype on ChatGPT hardly over, January 2023 saw 2 new GPT3.5 models: YouChat and Chatsonic. Both can access the internet and retrieve recent information, a feature that ChatGPT is still lacking. This means that,…
Read more → Something interesting has happened this week. How can ChatGPT return that Elon Musk is Twitter's CEO when that happened long after its training cut-off date? Let's start with my usual disclaimer: ChatGPT is a…
Read more → One thing that strikes me in the general communication on ChatGPT, is that surprisingly few people mention the fact that it's, in the end, one big word-string-generator on steroids (or a Stochastic Parrot, as Timnit…
Read more → In my previous articles I had a look at some core CxD components, and showed how you can let ChatGPT think it's a pretty good pizzabot. In this third installment, let's explore whether it can help me with NLP tasks,…
Read more → In my first encounters with ChatGPT, I had a look at that convo design classic: the welcome message. So now that we've passed that first hurdle, let's do what I always do when I test drive a new platform: build the…
Read more → If you've been online in the CxD community over the last few days, you've probably heard people talking about ChatGPT, OpenAI's newest Large Language Model. It's the youngest branch on the GPT family that brought us…
Read more → I get a lot of questions from aspiring conversation designers. This time: should convo designers know code? Here are some arguments for and against! Yes, it will make you a better designer I often coach new conversation…
Read more → I hope you enjoyed the European Chatbot and Conversational AI summit as much as I did! Here's my list of which talks I'm going to rewatch. Jeff Dalton, Keynote on the future of collaborative task assistants in the open…
Read more → When I just started working as a tech writer (a long, long time ago), I often found myself wondering how to do things. How to find a way in which my activities made most sense. And most of all…whether I did things right.…
Read more → So you've been a conversation designer for a while, and you feel pretty comfortable in your role. But you also love learning. So now that you've got all your convo design courses nailed, completed all the challenges on…
Read more → This week, it's been 2 years since I went to the Chamber of Commerce and registered Convocat as my own freelance company. 2 years…and what a ride it has been so far! While I usually share my learnings and insights into…
Read more → From today onwards, conversation designers, linguists, bot builders and language lovers can discuss their favorite topics on Clubhouse! OK, it took me a lot of FOMO and the actual purchase of an Apple device, but there…
Read more → Earlier on, I talked about youre-not-married-to-your-texts for what it is: extremely useful input for improving the quality of your content. In this article, I share a handy 5-point checklist that I use for exploring…
Read more → So you've spent hours crafting that one article. Shed blood, sweat and tears designing that one perfect conversational flow. You hand it over to your editor, your peer reviewer or your client…and they smash it to pieces.…
Read more → If, like me, you work for corporate clients a lot, chances are that you'll be part of a chatbot scrum team. And that you'll work in sprints: fixed periods of time in which the team commits to finishing a pre-planned…
Read more →Quickstart SSML les 2: de SSML-simulator.
Quickstart SSML les 5: moeilijke woorden.
Quickstart SSML les 3: de basiselementen.
Quickstart SSML les 4: SSML invoeren in Dialogflow.
Conversation design is a career on the rise! And I sure notice that in my inbox: I get a lot of questions on how to break into this great profession. As much as I'd love to answer all of your messages personally, there's…
Read more → Samen met Brielle Nickoloff ben ik vorige maand gestart met een nieuwe editie van Voicelunch: eentje speciaal voor taalkundigen, conversation designers en iedereen die iets heeft met taal! We hebben inmiddels twee…
Read more → Last @voicelunch with Brian Roemmele was without a doubt the best voicelunch I never attended. Fortunately, @afirstenberg tweeted notes & quotes, so I could still make some sketchnotes. At voicelunch, we don't record our…
Read more → Conversation design is een vak met toekomst! En dat merk ik aan mijn inbox. Ik krijg steeds vaker berichtjes van mensen die meer willen weten over dit prachtvak. Helaas heb ik niet de tijd om iedereen persoonlijk te…
Read more →Quickstart SSML les 1: intonatie.
Interesting Tweet last week: Blender, Facebook's much-hyped AI chatbot, is reported to have turned rogue and start spewing anti-semitic responses. The problem is there. When you ask certain questions, you get…
Read more →A presentation exploring my transition from technical writing to conversation design, discussing how communication professionals are evolving their roles in the digital landscape. I explain that conversation design goes beyond simple chatbot scripting or dialogue trees, involving understanding user needs, communication patterns, and designing interactions that feel natural and helpful. Technical writers are well-positioned for this transition because their core skills, clarity, user focus, and structured thinking, are exactly what conversation designers need. The shift is not about learning entirely new skills but about applying existing expertise in a new context, leveraging experience with documentation, user guides, and information architecture while expanding into dialogue design, tone and voice development, and interaction patterns. The future of technical communication is conversational, and the intersection of technical writing and conversation design represents how organizations will interact with their users in an increasingly digital world.
An intro to the Natural Conversation Framework for modeling conversational agents.
Ik zie op social media steeds meer berichten voorbijkomen over hoe online vergaderen ons uitput. Als oorzaak wordt dan vaak gekeken naar communicatie, zoals het gebrek aan lichaamstaal, het elkaar niet in de ogen kunnen…
Read more → Want to know what you can do to make your voice action sound more speech-like? Let's continue with part 2 of my understanding-speech-part-1 for conversation designers, in which I discuss backchannelling, discourse…
Read more → Now that Corona has large parts of the world firmly in its grip, more and more Corona chatbots are appearing. While these attempts are in itself laudable, effective use of chatbots as crisis communication tools might…
Read more → On international women's day, I'd like to point you to a composer of some of the most beautiful music for voice: Hildegard von Bingen. She lived in the 12th century, an era in which it was perfectly normal to send off a…
Read more → With voice assistants on the rise, you probably have heard of SSML as a way to make your voice assistant sound less robotic and more humanlike. To use SSML effectively, it really helps if you're familiar with some of the…
Read more → (Dit artikel publiceerde ik in juni 2019 op LinkedIn) KLM Bluebot, Billie van Bol.com, Appie van Albert Heijn, Buienradar, de Voice action van Schiphol, Mijn vader ik ben er ook nog van Ben…het merendeel van de…
Read more →No items match the current filters.
Links
This article demonstrates how Jev, a typed decision-making model from TypeSafe, can evaluate agent responses by turning natural language judgment into structured classification with confidence scores. Rather than parsing generated prose, Jev returns constrained answers with probability and confidence metadata that lets developers identify uncertain evaluations and set appropriate automation thresholds based on risk tolerance.
An Organizational Second Brain: Building an AI That Learns From Experts - Engineering at Meta engineering.fb.com Meta's engineering team describes an AI agent system that captures organizational expertise as structured, auditable knowledge separate from reasoning procedures. The system combines domain-specific knowledge architecture with a self-improvement loop that implements expert feedback without model retraining, creating persistent institutional memory that scales specialist knowledge across organizations.
Cohere Labs' analysis of modern LLM training pipelines reveals that cultural diversity is systematically lost during post-training, even when models support multiple languages. The authors argue that building truly global AI requires moving beyond multilinguality to explicit cultural representation in training data itself, as inference-time alignment techniques assume cultural knowledge is already present in the model.
As AI systems automate routine customer service tasks, job titles remain unchanged but the actual work transforms dramatically. Support workers now handle exceptions and quality control of AI outputs; QA analysts evaluate both human and AI-generated interactions; knowledge managers become responsible for training data that directly shapes AI responses. This shift concentrates difficult cognitive and emotional labor among humans while creating an entirely new operational layer requiring constant monitoring, calibration, and governance, yet most organizations remain structurally unprepared for this transformation.
Maryam Miradi presents a seven-step architecture for building closed-loop evaluations of multimodal AI agents at scale, specifically addressing the core challenge that AI can improve image quality while altering truth. The framework emphasizes faithful reproduction of content (matching ingredients, portions, and identity) through human ground truth anchors, multimodal routing, cross-modal verification, QA gates with retry logic, production drift monitoring, and iterative tuning before deployment.
This empirical study examines whether people can reliably distinguish between stories written by humans versus generated by AI systems. The research investigates the cognitive and perceptual barriers that affect our ability to detect machine-generated text in narrative contexts, with implications for trust, transparency, and how we evaluate authenticity in written communication.
Bureau for Listening Manifest 3 presents listening as a multifaceted artistic, social, and philosophical practice that extends beyond hearing to become a mode of being and relating to the world. The manifesto frames listening as a radical form of attunement, resistance, and connection that can transform perception, community, and understanding across everyday, artistic, and institutional contexts. It positions listening proposals as generative invitations to reorient how people engage with sound, silence, resonance, and each other.
Storybook is a frontend workshop for building, testing, and documenting UI components in isolation. It enables teams to develop components without entanglement in business logic or application infrastructure, and serves as a single source of truth for UI patterns that can be shared across teams and consumed by AI agents for automated code generation.
DAIR.AI's weekly survey synthesizes four key AI research directions: self-improving agents that evolve their own scaffolds and weights, metacognition frameworks that unify confidence and self-regulation in LLMs, the structural conditions required for meaningful routing in multi-model systems, and empirical critiques of harness evolution showing that simple search baselines often outperform supposedly optimized prompts. The curated selection positions these papers as navigational tools for builders implementing agent systems in production.
Solo journaling TTRPGs are a growing sub-genre of tabletop RPGs that enable single players to tell stories through writing and random elements, without requiring a group. These games use mechanics like dice rolls, prompts, and resource management to guide narrative development, with examples like Thousand Year Old Vampire demonstrating how constraints like memory limitations create compelling storytelling dynamics.
The article argues that the most effective workflow for creating skill.md files (structured instruction files for AI agents) combines AI-generated drafts with human refinement. It contrasts purely automated generation, which lacks nuance and product-specific judgment, with fully manual creation, which is time-consuming. The recommended hybrid approach leverages AI for structure and speed while relying on human expertise to add the contextual knowledge that makes AI agents actually effective.
Interactional AI is a research programme that applies Conversation Analysis to improve conversational AI models and user interfaces. The core argument is that interactive interfaces are not naturally intuitive unless they are grounded in the foundations of human social interaction, requiring design teams to ground conversational UIs in conversation analytic research methods and social science approaches.
This article examines the friction that arises when users shift from exploratory conversation with AI tools to precise task delegation. Through direct observation of seven people working with AI, the author identifies three interaction styles (collaborative, commanding, over-explainers) and reveals how conversational fluency in AI interfaces creates false expectations of shared understanding, leading users to accept suggestions that drift from their original intent.
This piece examines the iterative workflow patterns users develop when interacting with AI tools to achieve desired outcomes. It questions whether current AI-assisted workflows are effective or if they devolve into endless cycles of back-and-forth refinement, and points to Figma's exploration of AI-enabled product development as a case study in optimizing idea-to-product processes.
Eve Fairbanks argues that AI-generated writing betrays itself through a distinctive, "frictionless" quality that emerges from the absence of human struggle and revision. Rather than being merely a neutral tool like spell-check, AI writing bypasses the cognitive work through which humans develop meaningful ideas. The efficiency that makes AI appealing to writers creates a crisis of trust for readers who sense something crucial is missing.
Taguette is an open-source web-based tool for qualitative data analysis that allows researchers to upload documents, create hierarchical tag systems, and annotate text with tags and notes for later recall and organization. The tool addresses the practical need for structured, systematic approaches to analyzing and organizing qualitative research data.
The human brain is inherently lazy and tends to take cognitive shortcuts. This cognitive tendency affects how we interact with systems and interfaces, suggesting that interface design should account for human cognitive limitations and our tendency to conserve mental effort rather than fighting against these natural inclinations.
This LinkedIn post highlights UC Berkeley Law School's AI policy that restricts student use of generative AI for core legal reasoning tasks like brainstorming, drafting, revision, and exams. The post curates and endorses this policy as a model for responsible AI governance in higher education, emphasizing that cognitive skill development remains essential for quality legal practice and education.
This webpage documents Doug Engelbart's legendary 1968 demonstration at the Fall Joint Computer Conference, where he and his team showcased revolutionary interactive computing technologies including the mouse, video conferencing, and collaborative editing. The page serves as a portal to archived footage, retrospectives, and historical materials commemorating what became known as "The Mother of All Demos," emphasizing both the technical innovations presented and the ambitious vision of augmenting human intellect that motivated the work.
Faculty at three universities are using Hypothesis social annotation alongside generative AI tools to promote critical thinking and ethical technology use rather than banning or ignoring AI. By structuring assignments that require students to evaluate, critique, and analyze AI-generated content, these educators help students engage deeply with course material while developing essential skills for thoughtful AI use.
Ben Shneiderman presents eight foundational principles for interface design that have shaped HCI practice for three decades. The rules address consistency, universal usability, informative feedback, closure, error prevention, reversibility, user control, and cognitive load. These principles, derived from experience and continuously refined, provide a practical framework applicable across desktop, web, and mobile environments.
Adi Leviim argues that the chat box interface became the default for AI products not through thoughtful design but through shipping convenience: it was the fastest UI to build around LLMs' text-generation capabilities in 2022. The essay contends that this default choice sacrifices decades of established interface design principles, maximizing both the gulf of execution and gulf of evaluation, forcing users into iterative refinement loops when they should have direct manipulation and structured information surfaces.
Amelia Wattenberger argues that chat interfaces are poorly designed for most use cases because they lack affordances, force users to articulate context that should be built into the interface, and isolate responses without enabling iterative refinement. She advocates for UI patterns that expose parameters and constraints upfront rather than burying them in natural language prompts.
This article presents a comprehensive architectural framework for agentic AI systems in 2026, arguing that most AI failures stem from architectural problems rather than model quality. The author defines four canonical design patterns (Reflection, Tool Use, Planning, and Multi-Agent) and emphasizes that agentic AI represents a paradigm shift from monolithic systems to distributed, observable, and bounded agent architectures.
This article argues that 2026 marks a fundamental shift from generative AI (passive, read-only text generation) to agentic AI (active, autonomous systems that plan and execute work). The author contends that AI engineers must move beyond prompt engineering and embrace systems engineering, treating AI as operational infrastructure that performs end-to-end tasks rather than conversational oracles that simply generate responses.
This article argues that AI value comes from redesigning workflows rather than inserting AI tools into existing organizational structures. The author contends that traditional org charts show hierarchy but obscure how work actually flows, and proposes work charts as a management tool that makes workflows, decisions, handoffs, and accountability visible for effective AI transformation.
This article argues that Retrieval Augmented Generation (RAG) has become a strategic imperative for enterprises in 2026, addressing critical challenges like LLM hallucinations, outdated outputs, and high retraining costs. RAG bridges the gap between large language models and organizational knowledge by retrieving verified, real-time data at the moment of generation, ensuring outputs are accurate, compliant, and trustworthy without requiring constant model retraining.
This is a documentation hub for GitHub Copilot adoption and best practices at NAV (the Norwegian Labour and Welfare Administration). The site provides news updates, tools, guidelines, and usage statistics for AI-driven development, focusing on custom agents, skills, and organizational implementation of GitHub Copilot across teams.
Anthony Masure traces UCD from Xerox Star through Don Norman, arguing that design cannot be centered on anything. A world built on invisible technology is a world without experience.
Fieldwork for Future Ecologies (Onomatopee 225, 2022) argues art practice and art-based research can radically expand the concept of "fieldwork" beyond its conventional scientific framing, positioning creative and speculative methodologies as essential tools for environmental and climate inquiry.
Vincent Rump introduces the Tri-System Theory (Shaw & Nave, 2026), which extends Kahneman's dual-process model by proposing a third cognitive system — artificial cognition (AI) — that actively participates in human reasoning and decision-making. The post argues that humans increasingly defer to AI outputs with minimal critical reflection, a phenomenon the authors call "cognitive surrender," which improves decision quality when AI is correct but degrades it when AI is wrong — while trust remains consistently high either way. This raises urgent questions about human autonomy, expertise, and responsibility as cognition becomes a hybrid, human-AI process.
Frank A. Kalman, a formally trained journalist, argues that using AI as a collaborative drafting partner is not a betrayal of craft but an honest extension of a writer's editorial process. He frames AI as a responsive thinking partner whose "wrongness" is productive, pushing the writer to clarify what they actually mean. The piece is a transparent, process-level account of how an experienced writer uses voice memos, AI chat, and iterative editing to reach a finished essay, positioned against the performative outrage of writers who conflate the suffering of drafting with the value of writing itself.
Tactical Tech is a Berlin-based international non-profit that develops creative interventions, toolkits, and training programs to help individuals, communities, and educators critically understand the socio-political and environmental impacts of digital technologies. Its flagship announcement, "Supercharged Human?", targets teens and educators with ready-to-use AI literacy resources designed to foster critical engagement rather than passive adoption. The organisation frames its work around building civil society capacity to resist misinformation, digital extractivism, and uncritical AI use.
China's proposed AI law acknowledges AI-related human vulnerabilities and establishes contextual technical measures to prevent AI harms | Edition #264
A peer-reviewed conference paper from DRS 2022 that examines how metaphors shape designers' understanding of machine learning and AI systems, exploring both where common metaphors mislead and what qualities make metaphors genuinely useful for design thinking.
AI has collapsed centuries of human knowledge into a single point of failure. Discover Algorithmic Epistemic Injustice and why statistical probability isn't enough.
6 plugins that turn Claude Code into a full dev team in 5 minutes. Each one covers a different role; planning, design, code review, security, memory, and team coordination.
Ben Wigler highlights a paper on emotion circuits in LLMs: discovery and control.
AI tools can accelerate research, but using them to avoid the hard work of learning undermines the development of independent thinking.
Apple's hybrid batch-incremental knowledge graph platform. Key inspiration for building smarter connections in this garden.
