Denmark's Ministry of Digitization reportedly abandoned a pilot to replace Microsoft Office with the open-source Collabora suite less than six months after it began. Tom's Hardware reports: According to Danish publication Computerworld, the pilot project was launched in 2025 and involved 38 employees, all of whom would start using the British open-source Collabora office suite instead of Word, PowerPoint, and Excel. The experiment was initiated by the then Digitization Minister Caroline Stage, who wanted to reduce the country's dependence on American technology. Denmark isn't alone in moving away from U.S. software, with France and the Netherlands both taking steps to replace Microsoft Windows. However, it seemed that the experiment fizzled out less than six months after it started. It had reportedly ended by the time the Danish government called for an early election in February 2026, with some employees complaining about issues with style, paragraph formatting, blank pages, and challenges with tracking changes.
But even though it's been months since the experiment reportedly ended, Prof. Mikkel Flyberbom of the Copenhagen Business School complains that the project hasn't been properly evaluated. "It's a good idea to test alternatives in a pilot project," the professor said in an email. "However, I wish that the ministry would take the opportunity to learn from the pilot project and share those experiences with other parts of the public sector."
Internet Archive’s digital librarian, Brewster Kahle, leads a tour through the Physical Archive.
Visitors who came to the open house at the Physical Archive in Richmond, California, on October 20 were curious. They wanted to learn about the range of items collected, the process of preservation, and the staff responsible for making it all happen.
As people toured the space and observed the various media being saved, many also expressed support and reverence for the Internet Archive’s work to provide universal access to all knowledge.
“I really care about preservation and digital culture. Places like this are important to maintaining and building the society we want to live in,” said Sidney Le, a film archivist who now works in logistics, of his third visit to the Physical Archives. “It’s good to see that people here are taking seriously the idea of holding on to culture and preserving it.”
A Closer Look at Collections
Liz Rosenberg
Liz Rosenberg, donations manager for the Archive, welcomed the crowd in Richmond and provided directions for the evening.
“Thank you all so much for coming to our open house, where we can showcase the amazing preservation work done in the Internet Archive,” she said. “Here, you will see many special collections and interesting preservation demos showcasing the work that our incredible team does to preserve, digitize, and make accessible collections to patrons from all over the world.”
Rosenberg invited visitors on tours of the facility with Internet Digital Librarian and Founder Brewster Kahle, as well as a chance to see the Scribe book digitization workstation, a presentation on dehumidification of books, and observe microfiche digitization.
For visitors in the special collections room, Caslon Kahle pointed out new donations on display including Romanian history documents from the estate of professor Keith Hitchins, climate records from the National Oceanic and Atmospheric Administration (NOAA) and materials from more than 9,000 Freedom of Information Act requests filed by San Francisco’s legendary researcher Ernie Lazar.
Why These Collections Matter
Ivy Shaw and Max Buettner, both doctoral students at University of California, Berkeley, came to the event because they said they are big fans of the Internet Archive.
Buettner said he was impressed by the sheer quantity of material at the facility and unique items in special collections, such as a paleotectonic map of the Jurassic. Studying classics, he said he relies on the Internet Archive for rare materials he can’t find anywhere else – once discovering a manual on the Etruscan language online.
“Our library at Berkeley did not own a copy of this very important book, and the Internet Archive did,” Buettner said. “It’s sort of a miracle that we have this Internet Archive that can supplement these university libraries.”
Shaw, who is in a French PhD program, said seeing the books at the Physical Archive on so many topics made her appreciate the nonprofit organization’s effort to make knowledge available to all. “The Internet Archive allows you to access things in a very ethical way so you can feel good about supporting it,” Shaw said.
On a recent Friday tour of the Internet Archive’s headquarters in San Francisco, Shawdi Sani became curious about how the physical material was stored, so she registered for the event. Exploring the Richmond facility, she was captivated by the microfiche station and the customized software and equipment used in the restoration. “It was interesting how they brought that back to life,” she said. “It was kind of archaic and now it’s very useful.”
Sani, who works in applied machine learning, said she’s interested in seeing how the internet has evolved over time.
“As someone in Gen Z and with AI everywhere, there’s a huge need I see among my peers to go back to when the internet was less monetized and more raw,” said Sani, who enjoys exploring old websites, such as Gawker, saved by the Archive.
Preserving Today for Tomorrow
New to the event this year was a recording studio set up in an Airstream trailer parked outside the warehouse.
“Talk to the future,” Marc Weber said to visitors who walked up to check it out. Those who were game were asked to pull a question out of the hat such as “Why do you think preservation is important?” or “What’s your favorite item/collection on the Internet Archive?” to answer.
The soundbites will be shared on social media and preserved online. “It’s an informal way to capture what people are thinking today,” said Weber, an historian and curator for the Internet Archive. “Ten or 20 years from now it will be interesting to see what was on people’s minds.”
There were also postcards set out for people to fill out asking: “What’s your commitment to preservation?” Then, Archive staff will mail the postcards to them in three months as a reminder of their pledge, explained Chris Freeland, director of library services at the Internet Archive.
Charlie Robinson, an engineer from San Francisco, said he relies on the Internet Archive often when he can’t find a book or piece of media anywhere else. “The Internet Archive is one of those things where if it didn’t exist and somebody tried to explain it to you, it wouldn’t sound like an inevitable thing or even possible,” he said at the event in Richmond. “And I’m so glad that I live in a universe where it exists.”
The Internet Archive’s efforts to preserve all forms of media and chronicle digital history is something to be revered, said Eric Ma, applied scientist working in tech in the Bay area.
“It’s all in the public good and not motivated by profit. Honestly, I can’t really name another organization quite like it,” said Ma, who especially enjoyed seeing the collection of vinyl at the Physical Archive. “So many people are willing to donate their whole collection over to the Archive. It makes me have hope in humanity a little bit more.”
The following article originally appeared onThe Arize Blogand is being republished here with the author’s permission.
Every time you think you’ve just got the hang of this AI thing, another thing gets declared “dead.” This week, it’s agent harnesses.
Back in ancient history, aka June, I wrote a piece called “Frameworks are dying, harnesses are winning” and I don’t take that back. But this week a research team took a specialized agent harness, trained its behavior into a model and then took the harness away. The trained model handily beat the original model+harness combination. So are harnesses done now?
Betteridge’s Law applies, so the answer is “no.” But the job of the agent harness is changing. Parts of a harness that are sufficiently generalizable can be trained into the model. That leaves your harness holding the things that make your application unique: your tools, your data, your users, and your environment. You still have to build a harness, but you have to be prepared to regularly throw it away and rebuild it to keep up with evolving models.
This blog is about why that’s true, and how to turn “constantly rebuilding your harness” into an automatic process, not a giant time suck.
How Harness-Zero trained an agent harness into a model
First, the paper that kicked off this train of thought for me: it’s called “Harness-Zero,” from Haoran Ye and colleagues at Peking University. Its subject is harness distillation, i.e., training a model to reproduce behavior that used to be the result of the harness. The methodology is a little involved, so stick with me.
First, the team built a specialized harness for a small model, Qwen3.5-9B. An automated loop, run with Kimi K3, analyzed where the small model kept failing on training tasks and added the usual components to fix those failures: custom tools, middleware rules that block risky actions, skills files and memory. They called this the “evolved harness.”
Then they used that harness to generate training data. They put the small model through training tasks inside a minimal harness while a stronger model, GPT-5.6 Sol, reviewed every step the small model proposed. Sol used a copy of the evolved harness to tell it whether that step was a good idea. Sol passed good ideas from the small model through unchanged, and whenever the small model and the guidance disagreed, Sol made the smallest fix needed before the step ran.
That gave them a bunch of training runs and, crucially, data about what needed to be corrected. Then they created the distilled model: They fine-tuned the small model on the corrected runs. Then they removed the evolved harness, the reviewer, and everything else.
They then tested both the evolved harness and the distilled model head-to-head on spreadsheet editing, multi-app tool use, and chemistry. The distilled model’s average task success rose 21 points versus its untrained predecessor. That’s more than the untrained model gained with the full evolved harness attached. So the distilled model wasn’t just as good as the model+harness, it was noticeably better.
What harness distillation can and cannot move into the model
It’s not that the distilled model ran with no harness whatsoever: It still ran inside a minimal loop with one Bash tool, and the authors say that their method narrows what a harness has to provide rather than eliminating it entirely. The distilled model also didn’t win every competition: In chemistry, where the task is predicting which molecules combine to make a target compound, the distilled model trailed the harnessed version by eight points. The harness there supplied chemistry knowledge, candidate-generation logic, and a molecule validation tool. The authors’ explanation is that step-by-step procedures, like inspecting before editing and verifying before finishing, are much easier to train in than deep domain knowledge. Context management didn’t fully transfer either. So there are still some things harnesses definitely need to do.
Why training on corrected runs made the difference
But the most important finding, in my opinion, is that they showed what kind of training made the difference:
Fine-tuning on the stronger model’s own successful runs left the small model’s score exactly where it started. Just knowing somebody else is smarter isn’t enough.
So did fine-tuning on runs the small model completed inside the evolved harness. Again, all that teaches the model is “somebody is better than you at this.”
Giving the reviewer the correct answers produced nearly perfect training runs and a model that barely improved, this time because it overfit and learned nothing.
What worked was training on the corrected runs: The model was able to see runs that looked a lot like what it would have done, but with small changes still within its capacity.
Remember that part, because it’s going to be important.
Three training signals left the score flat or overfit: the stronger model’s own successes, runs completed inside the evolved harness, and a reviewer given the right answers. Training on corrected runs, where each fix was still within the small model’s reach, raised task success by 21 points.
Why models are absorbing parts of the agent harness
In Rich Sutton’s 2019 essay The Bitter Lesson, he looked back over 70 years of AI research and found the same pattern in chess, Go, speech recognition, and computer vision: Researchers built their own knowledge of the problem into their systems. That helped in the short term, then plateaued, and eventually lost by a wide margin to general methods, search and learning, that improve as computation gets cheaper.
A harness is that kind of built-in knowledge, and model training is the learning that ends up absorbing it. Sutton’s second point is also relevant: He argued we should build in “only the meta-methods that can find and capture this arbitrary complexity,” rather than the things we’ve already figured out. For harnesses that means: don’t just build it, but capture the process you used to build it, and automate that process.
But that’s a general prediction. More recently people have been saying very specifically that “models are going to eat the harness”:
In January, Jesus Rodriguez at TheSequence wrote “The Great Absorption,” arguing that hand-coded agent scaffolding keeps losing to capabilities models learn directly.
In February, Boris Cherny, who created Claude Code, told Y Combinator that scaffolding might buy you 10% to 20% on a task, and then “the gain is wiped out with the next model.”
In March, Vivek Trivedy at LangChain predicted that planning, self-verification, and long-running coherence would move from the harness into the model.
And so did lotsofother people, probably more than I’ve been able to find.
Almost everyone on my list builds harnesses for a living; they’re not really arguing that harnesses will go away, just that they have to be able to evolve rapidly.
What still belongs in an agent harness
As noted already, some types of knowledge don’t seem to transfer well from distillation. In ARC Prize’s results for ARC-AGI-3, Gemini 3.8 Flash scores 3.4-times higher with one harness than with another. The better harness keeps the model’s reasoning state between requests and compacts long conversations. That’s context management, one of the things Harness-Zero said they couldn’t fully train in.
Anthropic’s Opus 5.5 prompting guide, released this week, also shows evidence of shrinking harnesses. It suggests deleting “think carefully” instructions because the model now sets its own amount of thinking, and it tells you to retest scaffolding you built for charts and screenshots because the model reads them better without that help now. But at the same time, Anthropic added new harness pieces: a system prompt paragraph that keeps unattended agents from stopping early, a reminder your harness sends when a long run goes quiet, and elapsed-time budgets for multi-agent setups.
So harnesses aren’t going to shrink to nothing, but what’s in them is going to need to evolve with every new model generation, and the parts that survive are the ones tied to your specific tools, data, and environment.
How to evolve an agent harness as models improve
If your harness has to change every generation, the natural move is to automate the changes, but Google’s RRSI paper shows how that can go wrong. Unconstrained automatic harness evolution overfits, meaning the harness gets better at the tasks it’s tuned on without getting better at the general job. In their tests, the unconstrained version scored highest of any variant on the tasks it evolved against, and landed within a point of the original, unevolved harness on benchmarks it hadn’t seen.
RRSI’s fix is mostly bookkeeping: For every candidate change it records which component the change touches, what it was supposed to fix, the diff, how score and cost moved, and whether the change was kept. Components that stop producing gains get removed.
A practical agent harness maintenance loop
You don’t need an automated evolution loop to borrow that discipline. Here’s what it looks like by hand:
When a model ships, tag traces with the harness version and the model version, log what the harness changed, turn off one component at a time, and rerun evals. Keep what still helps and delete the rest, then repeat on the next model.
Tag every tracewith the harness version as well as the model version. When either one changes, you want to know which one moved your results, and reading the full traces tells you far more than the scores alone.
Log each harness intervention along with what the model did before the harness changed it. That’s data you have right now and are probably just throwing away. That’s what produced the record Harness-Zero trained on, and it’s the same shape Perplexity uses to train its Computer agent: learning from good steps in real sessions and correcting failed tool calls, which cut tool-call failures by 21% between its earlier and later trained versions.
When a new model ships, turn off each harness component one at a time and rerunyour evals. Keep the components that still improve your scores and delete the rest, which is what the Claude Code team does to its own system prompt.
So don’t delete your harness, but don’t treat any version of it as finished either. Keep the record of what it corrected, get rid of what the latest model already does without help, and put your effort into the parts that are really specific to your problem, because nobody else has the traces to train those into a model for you.
More and more of your terminal work now runs through tools that push the terminal harder than a shell prompt ever did: AI coding agents like Claude Code and Codex, full-screen apps like Neovim and tmux, system monitors and dashboards. These tools lean on modern terminal protocols, the kind today’s best standalone terminals are built on.
Starting with 2026.3, the built-in terminal in JetBrains IDEs speaks that same language. It now runs onlibghostty, the terminal emulation library that powers Ghostty itself.
Many projects already embed libghostty, and it’s widely regarded as a reliable, performant library, so we preferred it over the other options we considered.
In our IDEs, libghostty replaces JediTerm, the terminal emulator we relied on for many years. TUI apps increasingly depend on modern protocols that JediTerm never supported. That gap felt more like a minor inconvenience, since the experience didn’t quite match other terminals, but it wasn’t disruptive enough to warrant a bug report. Now libghostty takes over parsing and terminal state in our IDEs, while the IDE still handles rendering, and higher-level features and integrations.
Here is what those changes mean in practice.
Multi-line prompts and keybindings work the way the docs say
With support for the Kitty keyboard protocol, terminal apps can now distinguish among Enter, Shift+Enter, and Ctrl+Enter, and can request key-release and repeat events. CLI AI agents like Claude Code, Codex, and Pi, as well as the fish shell, and applications like Neovim, all use this protocol. The terminal also supports xterm’s modifyOtherKeys, an older keyboard protocol that Vim uses.
For you, this means Shift+Enterstarts a new line in your prompt instead of sending a half-written message, and shortcuts that combine Ctrl, Shift, or Alt with other keys reach the app intact. Terminal-based editors and tools respond the way they should, with no IDE settings to adjust. The old setting that sent Esc+CR on Shift+Enter is no longer needed and is ignored.
Your tools follow your IDE theme
Switching the IDE to Dark mode used to leave terminal apps guessing, and you ended up with dark text on a dark background or a bright color scheme inside a dark IDE, unless you configured each tool by hand.
Now the terminal answers the dark/light query (CSI ? 996 n) from the current IDE theme and sends a notification through mode 2031 when you switch. Claude Code, OpenCode, Pi, Neovim, and tmux subscribe to these theme-change notifications, so most of them switch automatically with the IDE.
The terminal also answers OSC 4 palette queries and the Kitty OSC 21 color query, in addition to the OSC 10/11 queries that JediTerm already handled. OpenCode and Pi read the palette, and Junie CLI asks whether the terminal is dark or light. Programs can now set the extended palette colors 16 to 255 with OSC 4, while colors 0 to 15 stay under the control of the IDE theme.
Terminal applications with rich output now draw better
AI agents, system monitors, file managers, and dashboards are only as good as the terminal that renders them. The new engine handles almost 50 control sequences that JediTerm either ignored or misread. A few you may have hit:
REP (CSI Ps b): ncurses uses it for runs of 7 or more identical characters, so some lines previously rendered with missing characters.
Colon-separated SGR colors (38:2::r:g:b): JediTerm printed the colons and merged the digits into the wrong attribute.
Unknown APC, SOS, and PM strings: JediTerm printed them as text. They are now silently dropped.
Also supported are left/right margins (DECSLRM), protected areas, tab-stop control, insert mode, XTSAVE/XTRESTORE, the Kitty CSI 22 J scroll-and-clear, and X10 mouse reporting. Cursor blink mode (?12) is respected now; before, the cursor always blinked. There’s much more – to see the full list of supported escape-sequences, please refer to this table.
In summation: lines, borders, and separators no longer come out broken, colored output no longer shows stray colons or the wrong colors, and leftover invisible codes no longer show up as garbage text.
And this is just the start
The libghostty library already parses many sequences that the IDE doesn’t act on yet: OSC 52 clipboard, desktop notifications (OSC 9/777), ConEmu progress bars, Kitty graphics, and styled and colored underlines. Parsing is no longer the bottleneck. Only the IDE side is left, so expect these to arrive faster than they did with JediTerm.
In practice, that groundwork covers features you may know from standalone terminals: copying to your clipboard from a program running on a remote server, getting notified when the app in the terminal finishes a task (for example, Claude finishes processing a prompt and is waiting on you), seeing a progress bar for long tasks, and showing images directly in the terminal.
Try it now, enabled by default in 2026.3
The libghostty-based engine is already on by default in the reworked terminal in the 2026.3 EAP builds of your IDEs, and it will stay that way through the Beta and Release Candidate builds and in the stable release.
If something goes wrong, you can switch back to JediTerm: go to Help | Find Action, open Registry…, and turn off terminal.use.ghostty.emulator. We’d appreciate it if you also let us know what happened by leaving a comment on IJPL-218849 or by filing a bug report.
Thank you, Ghostty
This story began with the Ghostty team, who approached us at a conference last year and asked whether we would consider using their emulator at the IntelliJ IDEA Conf. Since then, they stayed engaged through every discussion that followed. Many thanks to Mitchell Hashimoto, Tim Culverhouse, and all the Ghostty contributors. You built libghostty as a reusable library and kept improving it while we integrated it. Over the past months, libghostty also gained features that JediTerm users relied on, which made the switch possible. We’re glad to build on your work, and we look forward to contributing back.
Kaloyan Chalakov: The Agile Team That Built Great Architecture But Forgot the Customer
Read the full Show Notes and search through the world's largest audio library on Agile and Scrum directly on the Scrum Master Toolbox Podcast website: http://bit.ly/SMTP_ShowNotes.
"You don't have to build it only because we can. We need to lead to a purpose." - Kaloyan Chalakov
Kaloyan worked with a technically strong team that could build impressive architecture, pipelines, and infrastructure fast. Sprint after sprint, they showed the quality of the technical system they were creating. But the hard question stayed unanswered: what customer problem were they solving? The wake-up moment came in a sprint review, when a client finally joined and told the team that the network and architecture were nice, but they did not solve the actual problem. The customer needed one small piece of functionality, and the team could not point to it. The room went silent. The Product Owner felt exposed, the team felt embarrassed, and Kaloyan saw the moment as painful but necessary. Because the mistake happened early, it became a learning opportunity instead of a product disaster. The team learned to connect technical excellence to product value, and to ask sooner whether their work solved the customer's real need.
Self-reflection Question: What are you building because it is technically interesting, and how will you test whether it solves a real customer problem?
Featured Book of the Week: Start With Why by Simon Sinek
Kaloyan recommends Start With Why by Simon Sinek because it changed how he thinks before acting. The book pushed him to pause before solving a problem and ask what the real purpose is. He started applying that question in workshops, team conversations, and product discussions. When a team member asked why a leadership workshop included a specific activity, Kaloyan realized his explanation had not landed. Instead of pushing forward, he stopped, reset, and rebuilt the workshop around a clearer purpose. For Kaloyan, the "why" is not a one-time statement. It is a question to ask at every step, because people do better work when they understand how their part connects to the bigger goal.
[The Scrum Master Toolbox Podcast Recommends]
🔥In the ruthless world of fintech, success isn't just about innovation—it's about coaching!🔥
Angela thought she was just there to coach a team. But now, she's caught in the middle of a corporate espionage drama that could make or break the future of digital banking. Can she help the team regain their mojo and outwit their rivals, or will the competition crush their ambitions? As alliances shift and the pressure builds, one thing becomes clear: this isn't just about the product—it's about the people.
🚨 Will Angela's coaching be enough? Find out in Shift: From Product to People—the gripping story of high-stakes innovation and corporate intrigue.
Kaloyan Chalakov spent years making high-stakes decisions at sea before moving into Agile and organisational change, bringing lessons from safety-critical systems into how modern teams work.
1232. This week, we look at four ordinary words — "hippocampus," "chimera," "behemoth," and "gargantuan" — that all started as the names of monsters. Then, we look at where sailors got the phrase "graveyard watch" and at the difference between a graveyard and a cemetery.
The monsters segment was written by Samantha Enslen, who runs DragonflyEditorial. You can find her atdragonflyeditorial.com.
A hearty thank-you to our Keepers of the Commas on Patreon. We appreciate your support!
Larry Rosenblum
Mahala Russell
George Wilson
Laurel Paul
Linda Cox
Birna Anna Björnsdóttir
Kate Wade
🔗 Share your familect recording in Speakpipe or by leaving a voicemail at 833-214-GIRL (833-214-4475)
Log in
