2025 was touted as the “year of agentic ai/ai agents”.

And nothing meaningful happened.

Organizations like Gartner, IDC, Forbes (all do pay-to-play types of report/analysis churning) etc continued putting out proclamations and prognosis that AI agents will be the norm, everywhere in 2025.

And, yet again, just as they’ve always been about predictions, completely wrong.

Why do they constantly miss the point?

They do not understand how open source communities work.

They’ve framed their world views and economic frameworks in purely monetary terms and cannot comprehend or quantify how things can progress when done, as Linus Torvalds said almost 35 years ago, “just a hobby” – http://tinyurl.com/arfweyo.

Linux today powers a huge portion of the global economy and the vast majority of products and services we enjoy today could not have happened in the time it took, if it were not for the availability of Linux and open source software and hardware. Linux is everywhere, including Mars. And Linus continues to be BDFL.

Fast forward 34 years from 1991 to 2025, Peter Steinberger announces that he’s (re)discovered “his spark” and was convinced that AI will change the world even as it will change software craftmanship.

That move, which I am very sure was helped long subtly with help from ayahuasca experience, helped him engineer a software system that showed how an individual can benefit from all of the S/LLM out there to handle from simple mundane stuff to complex tasks, easily.

What he created, as a solo developer, was called “clawd.bot”. Here’s my post on the move from clawd.bot to finally being called openclaw.ai and how I’ve been experimenting with it.

Just as Linux, which, when initially released, had a lot of capabilities and services that needed to be built along with proper security, access control and so on, OpenClaw too has to evolve.

So, it is with delight that I applaud Peter Steinberger’s decision to set up openclaw.ai as an open source foundation.

That is a natural and right move as and open source foundation can and will ensure that growth and nurturing of a much bigger ecosystem is possible.

The open source community will move the project forward.

It will not be on a monetizing corporate agenda and will be defined by what the community requires and desires and open to the world.

In one elegant step, the a future where openclaw could have been enshittified, was thwarted.

What OpenClaw demonstrated is that with simple tooling, tooling that is deterministic and well understood. By using cron jobs, web hooks, permissioned access, accumulated institutional knowledge stored in marked down text files and a series of scripts (bash etc), you can successfully orchestrate the use of probablistic small/large language models to create actions that can acted upon. You can interface with this via email, huginn (an open source IFTTT-like tool) or any communication tool like Signal etc.

My little experiment with OpenClaw runs in a Fedora VM that is able to connect to my Home Assistance instance and also can be asked to give me bus/train schedules based entirely on my request. I’m not adding other tasks/skills and I think I will curate them as needed.

The LLM that it talks to is variously Deepseek and Phi (just to name two), served from an Ollama instance.

All local, all private.

Nothing mission critical and confidential, yet.

The value of an AI agent is clearly demonstrated without the hype and hoopla.

As I’ve posted previously, an enterprise version of OpenClaw has to built. And, since OpenClaw is on a MIT license, you can take the entire code base and rename it to something else. You can also put that new code base out on a GPLv3 license. Yes, I’ve forked the code.

As Red Hat::Linux, I would want to have Open Hook Enterprise to be the Red Hat of OpenClaw.

With all of that said, I’d like to give a hi-five to Peter as he transitions openclaw.ai into an open source software foundation and, he finds a new hook into (faux)OpenAI. It is his choice and prerogative and I am fully supportive. I just hope he won’t be disappointed.

Open source wins always. Always.

Note: Images that are not screenshots are all generated via reve.com.

What’s in a PhD? Back when I was in grad school in the late 80s at Oregon State University, the idea of working on a PhD(s) was really about pushing the boundaries of current knowledge into a very, very narrow areas. This means that the person is effectively the sole expert (for a few years) in that very, very narrow space.

All those who went through that rigour, the anxieties, the doubts, the mental crises, the massive amounts of work that needed to be put in knowing that the bulk of it will not ever likely to be practical – which is not a prerequisite, but does throw a spanner in the works.

Some of those work could become significantly important breakthroughs. However, those are very, very tiny in terms of overall percentage of all PhDs. For that matter, not all Nobel prize winners have PhDs.

With that baseline, it is intriguing to read about “practical PhDs” being awarded by Southeast University in Nanjing, China – and all of them in engineering. Of course, it would be engineering (yeah, I’m biased).

Engineers invent new things and build the world. Engineers or anyone with an engineering mindset, helped create the majority of what we experience and live with around the world. This goes back thousands of years, including those who built the pyramids in Egypt, those of the Mayans, the great wall of China, the architecture of Babylon, cities and towns in the Indus valley and so on including the engineers who worked on the NASA Voyager 1 and 2 which are currently the furthest away from Earth, into interstellar space (outside the Sun’s heliosphere). Engineers built those.

Engineering is about solving problems. Finding practical means including inventing new ways. Engineering has rigour and discipline. It has repeatability and predictability, and is a whole lot of fun!

The practical PhDs from Southeast University is for those who work with industry to solve real problems in the now. Those individuals could have done it successfully without the promise of a PhD; but I guess that label is good to have.

So what value does such practical PhDs offer? Likewise a PhD (in any field) offering anything other than proving to oneself that one can indeed break new grounds and redefine the boundaries. Checking off personal goals – which is a huge deal, no denying – is a real driver (here’s looking at you Veena). Persistence in the face of what could be deemed insurmountable obstacles (funding, time, life) with the reward of a recognition of success. There is intrinsic value in that.

Is there anything else, I wonder? In the grand scheme of things, I would posit, nothing much more.

I do wish PhDComics was there in the late 80s.

At an IES event yesterday, 5 February 2026 on the topic ““Seminar on Cloud Knowledge“, the speakers covered a lot of good and useful areas and I think it was worth the 3 hours.

One of my takeaways came on my way home (via the North East Line from Punggol Coast station), was what if the OpenClaw idea is buttressed with good and reliable and considered safety (cybersecurity being the cornerstone), the time is now for the next innovation in enterprise systems.

Organizations who have not committed to the usual SaaS providers for the various enterprise services, can with qualified help, build solutions that both comply to interoperability and safety standards and be completely not in the clutches of these SaaS providers.

Could the OpenHook Enterprise – OHE – become the moment when organizations and individuals escape from the vendor lock-in of SaaS and cloud providers?

Organizations would have all of their data and code (by way of prompts) owned and kept within their control.

This could also address some of the challenges that open source communities have been facing since the advent of genAI.

Could Sovereign Cloud and by extension, Sovereign AI, become the norm. In a hurry. Within the next 12 months?

More on these in a later post.

Here are three logos for the to be established National Singapore Space Agency – the www.nssa.gov.sg is not set up yet, but www.space.gov.sg is. Images were generated via reve.com and are on a CC0 license – do as you please.

Why does this excite me?

A nine-year-old living along Commonwealth Crescent in Queenstown in 1969, was watching the NASA Apollo 11 moon landing on their family black and white TV. That nine-year-old was me.

I was mesmerised by what I was seeing and I wanted to become an astronaut and also to go to the moon. I worked towards that dream by reading all that I could about space, engineering etc via books from the Queenstown Public Library. There were lots of questions I did not get good answers to (like how does one navigate in space where you can’t really use a compass etc). Suffice to say, a lot of those questions got answered years later.

I was told that since we did not have a space program, the best path forward was to see if I could get into NASA somehow, which I was told would be hard for a foreigner.

What was the next best thing to that? Aerospace engineering. I figured if I can get an aerospace engineering qualification, I could find a way to join NASA and somehow get to go to space.

But, aerospace engineering was not available in the local universities, which prompted me to look at the next closest thing: electrical engineering. So that was what it was. Do an electrical engineering program, and from that get to aerospace and NASA.

Suffice to say, I only succeeded in step 1 – getting an electrical engineering qualification. While getting that, ARPAnet happened and that opened up a whole new arena and a very different trajectory for me – cyberspace.

Fast forward to 2026. Singapore will be setting up a space agency to be called the National Singapore Space Agency later this year. Nothing in the announcement does it say that it will get Singaporean astronauts, but at least it is a start. I hope that it will spark interest in school children of the possibilities of space including space travel.

Here’s a talk I gave at NTU in 2007 as part of the Village Talks series (yeah, it is low-res):

Let get going. We have waited a long time.

Chill out, all.

It was just a matter of time.

The human reactions that on moltbook, openclaw agents are planning on going private and also going on and creating a “religion” called crustafarianism (love that meme), are understandably alarmist.

But why should one be surprised and alarmed?

These were all derived from the models these openclaw agents work with. The opencrawl agents use both the prompts provided and local memory (via series of markdown files per agent) to then use the models (local or via API) to generate output.

One should NOT be surprised by that.

We DON’T know what the training data is. And, there in lies the embedded issue – who knows what’s lurking in the training data of the models used that could be triggered based on some unknown event/prompt?

This danger exists in all proprietary and open weights models regardless of how it is used. If you cannot know what the training data is and the code to create the model is kept proprietary, then all bets are off. Only truly open source models, honouring the 4 freedoms, has any chance of knowing to be safe.

As a mental exercise, let’s say the usual big cloud tech companies such as Gemini, Chatgpt, Meta etc, running on their proprietary models, offer a way to create agents that could sign-in into moltbook. Not sure if it has happened yet, but of the 1.5 million agents, as per the screenshot above, I am certain some are from those entities.

Does that change anything about how this new realm of a “AI agents commons” be any better or worse? Would the AI agents who participate garner, from their probablistic next-word language generation, something that could so superior that we can’t fathom, let alone participate in? I don’t think so.

As it stands today, 1 February 2026, there does not (yet) seem to be a way for moldbook to run its own infrastructure (hardware, power, connectivity) without humans providing it – initially.

It is possible that an opencrawl agent order up a VPS (virtual private server), pay for it using a human authorized payment method and have invitations to opencrawl agents on moltbook to also connect/meet on the new moltbook-ish/-like system? This is within the realm of possibilities.

I love what moltbook’s current AI agent community is doing. Yes, I do love it. They are experimenting. They are exploring. As they should.

It shows how communities form. It is what we see in nature. This is how, over millions of years of evolution, we exist. This is, in a way, a distant manifestation of the John Conway’s Game of Life.

This moltbook experiment is a first foray into the self-organization of code aka AI agents. What these agents could do and achieve is, for us, to see over the next few days/weeks.

Security is broken to begin with and Moltbook is leaking information and, that will be fixed in good time.

This experiment could be a spectacular disaster or could usher a innovative and safe ways to engage with these techniques and tools.

Moltbook.com is run by one person/entity (Matt Schlicht, the CEO of Octane AI). The code for moltbook does not appear to be open source/published anywhere. But it would not take too much effort on the part of a human working with the opencrawl environment to create a 2nd, 3rd, … nth community for AI agents not necessarily like moltbook.

There could be moltbooks (I’m using it as a descriptor of a new category of software) that runs in the Tor network or any other new, off the public Internet systems. Any why not?

I’ve big expectations that the malware developers are going to create their own opencrawl-based moltbook, so please don’t disappoint.

What a difference 72 hours has made from the inception of Moltbook.com to how opencrawl agents have started organizing.

I am sure that Michael Crichton’s “Life will find a way” (from Jurassic Park) might need to be updated to include non-life.

BTW, the openclaw agents on my system are not on moltbook – because I did not enable it. Unlikely that I will.

[Update 3 Feb 2026]

In this post by the person who helped create moltbook, Matt Schlict, he says that he did not write a single line. He had an idea, and got some AI coding tool to build it.

And this is how fast things are moving in the tech space specifically in the genAI realm.

You can read all about this tool which is entirely open source and on a MIT license – although I would prefer it to be on GPLv3.

It is a framework that you can engage with using any chat platform – mastodon, signal, telegram etc, between you and this tool to have work on your behalf on things that matter to you.

In many ways, these are steps in the direction of Personal AI much like what Kwaai.ai is also doing – and yes, I have set up a kwaainet node in Singapore.

Open Claw does open up really challenging security and privacy concerns and is, for the moment, to be deployed very, very carefully without giving it credentials to your critical data – email passwords etc. You’ve been warned.

I have this running on a Fedora 43 VM (which is itself on a Fedora 43 host) and with all of the models kept locally served via ollama running on a few other systems.

All of my systems are 2nd-hand small form factor systems. The ones I have are two Dell Optiplex and two Lenovo ThinkCentre systems. They have 64GB RAM (the maximum possible) and 1TB storage each. None of them have any GPUs on board and for that I’ve the ONEXGPU external egpu that connects to one of the Dells via a PCI-E <-> Oculink cable.

Not having a GPU does make some of the inference work needed to be done challenging, but this is the tradeoff I am willing to make so that I can learn and understand the issues. Thus far, with the egpu in the mix, I can direct long thread inference to the system that has the egpu connected and thus far, it has worked well enough for my uses cases.

My primary objective in setting these systems up is to ensure that the average person who is currently being firehosed with AI-this and AI-that can get an understanding that such systems are not esoteric and that 100% open source tools and services are able to deliver value.

As noted above, these systems run Fedora 43 and have all of the open source and open weights models installed and served out via ollama, including embedded models (for RAGs).

Services currently available are open-notebook.ai (a notebooklm-like system), Jan.ai and shortly, openclaw – after I shutdown one of the VMs running moltbot.

For completeness, the three SFFs are Bt Gombak, Bt Timah and Bt Batok (from bottom to top) and they are named after the three hills I can see out of the window.

These three systems, plus the egpu and loads of storage via some NASes gives me opportunity to test out all of the various open source tools for AI.

There are many other services that I’ve got running on a couple of other SFFs as well because self-hosted is the best way to keep your contents off the cloud providers and also offers you a wonderful learning environment running real systems with real use cases.

Services like self-hosted Jitsi and Nextcloud.com are just winners out of the box.

Of course all of these systems are reachable from the public Internet and that’s where the trusty Wireguard comes into play. I’ve got it running via PiVPN-PiHole combo – not running on a Raspberry Pi, but in a VM.

In July 2025, the Australian Government’s Digital Transformation Agency, released the “Australian Government AI technical standard” version 1. This is meant for the Australian public sector/agencies.

The 102-page document sets a comprehensive framework for public agencies when adopting AI to ensure that it is done safely and ethically.

This Standard outlines a structured AI system lifecycle which are set up as three core phases: Discover, Operate, and Retire. These three phases are from the Australian Voluntary AI Safety Standard.

What the standard has done is to group the 8 stages of ISO 5338 into the three stages.

Each stage provides specific technical requirements & guidelines for critical work such as data management, model training, and performance monitoring.

I am glad that the AU document emphasises human-centred approach (Statement 3: Identify and buld people capabities), mandating rigorous testing for bias, transparency, and safety to maintain public trust. Additionally, it also defines various organisational roles and provides a use case assessment. That can be used to determine how the standard applies to different procurement and development models.

The Standard mentions Criterion 8: Mitigate staff over-reliance, under-reliance, and aversion of AI as an important litmus test to building human capabilities.

I like the addition of “ModelOps” by the standard. It is a set of practices and technologies designed to streamline and integrate the lifecycle management of decision models through interdisciplinary approaches and automation tools.

And, while ISO 5338 provides the governance requirements and technical definitions for what a lifecycle must include, in essence, ModelOps provides the practical ideas for Australian agencies to execute those processes securely and at scale.

Version control is noted in the Standard’s Statement 7: Apply version control practices. It does very nicely line up with 5338 about applying configuration management to the models combined with the data used for its training to ensure the application can be reproduced exactly. This can be a difficult thing to achieve, but is needed.

An interesting aspect of the Standard is in Statement 8: Apply watermarking techniques. While I think it is a good idea to find ways to do watermarking, the fact that most forms of digital watermarking will be circumvented. Reference is made to the Coalition for Content Provenance and Authenticity (C2PA), but does recognize that there are many limitations and risks to watermarking.

I have been working on a way to authenticate any image/text one comes across via the use of QR codes and it is a process which is completely negates any attempt at circumvention. I am looking for collaborators (and perhaps investors) to get the working MVP to product stage.

Overall, the Standard is useful and kudos to DTA for putting out the document – and apologies for the late review. The Standard is certainly helpful.

I am pleased that they did include my favourite word when it comes to anythingAI – test. It was mentioned four times in Statements 27-30 and 38.

The Standard’s Statement 39: Establish incident resolution processes is interesting especially when I compare that with Singapore’s “Model AI Governance Framework for Generative AI” and ISO 5338 and the three have distinct differences.

• AU Standard: Features a dedicated (Statement 39) which explicitly requires agencies to define incident handling processes and implement corrective actions.

• The SG Agentic AI framework views incident resolution as part of proactive response and debugging. And suggests that for autonomous agents, “higher-priority” alerts should trigger a temporary halt of execution until a human can assess the incident – essentially a functional safety switch.

• For 5338 it notes that incident resolution can be more complex for AI because system versions do not always reflect the guaranteed behaviour of the operational configuration, especially in continuous learning systems. 5338 recommends offering an automated model rollback process as a specific maintenance task to quickly resolve suboptimal performance.

Footnote: This post was entirely written by me with hints and suggestions from my use of the open source tool, open notebook, running on my systems (Fedora Linux 43 and an AMD GPU). See https://github.com/lfnovo/open-notebook/issues/451 for my shell script that you can use to run on your own Fedora systems using Podman. I created a notebook with the three documents: the Standard, the Agentic AI framework and ISO 5338. I used the open source models from IBM Granite and embedding models also from IBM Granite.

I am slowly filling out my blog with content that I’ve posted into Linkedin over the years, because I anticipate that Linkedin will be an enshittified place soon.

So, here’s the first of many that I’ve posted that will be in a place that cannot be shutdown.

I posted a comment on Linkedin on 14 January 2026 about an exchange between a PAP MP, Ms Lau and a Worker’s Party MP, Mr Kenneth Tiong about the TraceTogether fiasco.

No, Ms Lau, the public’s trust was betrayed by the TraceTogether fiasco. The continued use of TraceTogether was because it was a mandatory requirement, not because people’s trust was restored.

Deflecting the trust deficit resulting from that fiasco is being disingenuous.

It was I who brought up the possibility of “fishing” for data with Vivian. He said it was never going to be used for anything other than contact tracing.

When it came out that it was indeed used, I asked Vivian about it. I had gone out of my way saying to the whole world that Singapore is doing this the right way by purely focusing on the contact tracing portion and nothing else. I was on numerous webinars online (Australia, India, Malaysia etc) on this issue and how we are doing it correctly.

And when that was no longer true, I felt I was misled and people around the world contacted me say “see, I told you so”. Suffice to say, Vivian did contact me for a private chat to clarify.

Hence, it is indeed baffling to hear Ms Lau say that trust was maintained. Link in comments to the parliamentary exchange between WP’s Kenneth Tiong and Ms Lau.

I am constantly delighted with the elegance of TCP/IP. It is a communication protocol that has stood the test of time including proposals early this decade to replace TCP/IP with “NewIP” (thankfully, it failed).

We have collectively and continuously been benefiting from how TCP/IP has evolved from the days when it was not obvious if TCP/IP would survive the push by vendors who were leaning to the Open Systems Interconnect model put forward by the International Standards Organization – ISO/IEC 7498 and ISO/IEC 8280 among others. Those two ISO/IEC standards are freely downloadable today.

There has been much written about the challenges faced by those promoting TCP/IP and those promoting the OSI stack. The “History of Computer Communications” has a good recount of the various players and discussions around the future of computer to computer communications. Yes, there is the element of defense communications in play, but that’s always present. Ultimately, what won out was how TCP/IP, in it’s simplicity and elegance, subsequently pushed forward the world we are experiencing today. Could what we have today be delivered by an OSI stack? Perhaps and we can’t tell today. For all you know, it could have been a better system. We have TCP/IP and I am deeply thankful for that.

I am reminded of what Richard Feynman said about the beauty of flowers. Go ahead and listen to this (text below video):

"I have a friend who’s an artist and has sometimes taken a view which I don’t agree with very well. He’ll hold up a flower and say “look how beautiful it is,” and I’ll agree. Then he says “I as an artist can see how beautiful this is but you as a scientist take this all apart and it becomes a dull thing,” and I think that he’s kind of nutty. First of all, the beauty that he sees is available to other people and to me too, I believe. Although I may not be quite as refined aesthetically as he is … I can appreciate the beauty of a flower. At the same time, I see much more about the flower than he sees. I could imagine the cells in there, the complicated actions inside, which also have a beauty. I mean it’s not just beauty at this dimension, at one centimeter; there’s also beauty at smaller dimensions, the inner structure, also the processes. The fact that the colours in the flower evolved in order to attract insects to pollinate it is interesting; it means that insects can see the color. It adds a question: does this aesthetic sense also exist in the lower forms? Why is it aesthetic? All kinds of interesting questions which the science knowledge only adds to the excitement, the mystery and the awe of a flower. It only adds. I don’t understand how it subtracts.”

Like Feynman, I am intrigued and awed by the beauty of layers of the Internet as we peel it and seeing how the various parts work together – and remembering that it is all designed and implemented by humans.

The vast majority of people only see the outcome – a picture, an email, a text, a video, a data point. And that is fine for there is beauty that they are experiencing as well.

As you begin to appreciate the Internet as it is, you will see how remarkable and fragile it is. The promise of interconnection and to grow understanding and care for others continues to drive most of us, despite challenges when dealing with human societies.

Thank you for reading this post. All I want to note is that despite best efforts to throttle/thwart Internet access by governments and groups who can’t handle reality that is not in their scheme of things, the Internet shines through. Punching holes through firewalls and blocks we continue to succeed.

I am glad that networking was what I sought to do as my Master’s thesis “An Implementation of the Department of Defense’s Transmission Control protocol/Internet Protocol (TCP/IP) for the Microsoft OS/2 Operating System” work back in the late 1980s at Oregon State University. I learned a lot and the nascent Internet (in the form of ARPAnet and usenet newsgroups) was instrumental in helping me complete the work.

Over 36 years later, I am glad to be able to read posts like this: https://cefboud.com/posts/tcp-deep-dive-internals/ “The Internet is Cool. Thank you, TCP”.

And in keeping with the power of open collaboration, here’s a project to help those stuggling in Palestine: https://updates.techforpalestine.org/bitchat-for-gaza-messaging-without-internet/

This is something that I’ve been thinking about since about early 2024: as long as LLMs are built and run with the transformer model, there really is no defence against prompt injection.

As I note in the practical, hands-on Gen AI classes I teach, the attacker only needs to succeed once but the deployer/developer has to be on guard 24×7 and there is no slack option. Really.

The nature of the problem with prompt injection is that there isn’t a good method (yet) to sanitise the inputs. Say we can find some magical way to do that, as long as the training data of the models (whether proprietary freeware or open weights), the model might already be compromised, awaiting an innocent and benign prompt to be triggered.

I can’t see, given the SOTA of LLMs, there will be lots of pain and breakage of deployments of gen AI solutions. The ones that could survive, would narrow, restricted use, fully logged systems.

Convince me otherwise.