Most of my posts are developed from something. However, this post is specifically developed from two sources. The first is an article by Eleanor Drage. For me, she is of unknown age, but looks under 40, putting her birth year to sometime after 1986. She is a senior research fellow at the University of Cambridge’s Leverhulme Centre for the Future of Intelligence, author of What If We Got AI Right?: How to Stop Catastrophising and Build an Ethical Future (2026) and co-author with Kerry McInerney of The Good Robot: Why Technology Needs Feminism (2024). The second source is a video by Andreas Spieß, who admitted to being 61, eight years ago. That would make him 69 now, which indicates a birth year of about 1957.
Canadian technologist and engineer Ursula Franklin (1921 – 2016) said, fantasies of technology would have it that innovation is always “investment-driven, shiny, lab-born, experimental, exciting”. In the real world, it is often “needs-driven, scrappy, on location, iterative, practical, mundane”. Pioneering technologies are usually useful systems = frugal tech, brought to life by people doing more with less. They don’t impose top-down solutions that complicate lives while making a few people very rich. Genuinely innovative technology set people free.
On 2025-08-02, Eleanor Drage wrote in the Guardian about The Association for Progressive Communications (APC) at Berlin’s once hippy, now increasingly corporatised Re:publica conference. There, she met researchers who are using technologies such as software-defined radios and spectrum sensing to allow people in low-resource environments to stay connected despite limited bandwidth, power, hardware and communication infrastructure. These technologies are the basis of the local community networks that supply coverage to the 2.5 billion people globally who lack internet access. In the Niger Delta, which suffers from toxic levels of air pollution from its oil industry, APC is setting up connections and deploying low-cost sensors that monitor the environment. These play a crucial role in how locals can advise children when to stay inside and which areas to avoid playing in. This infrastructure is managed for and by the municipality, serves a pressing need and can be installed and built by the people who deploy it.
Google Gemini and GPT are irrelevant to those who don’t have enough internet bandwidth to use them. The new digital divide is the gap between the top end of the global population that includes all readers of this weblog post – who have access to these power-intensive technologies – and those at the bottom, with little or no internet access. That is why people have to work out how to manage the trade-off between internet range and bandwidth, taking into consideration obstacles such as mountains and foliage.
Good innovation also often involves lobbying for good. Big Tech works in the opposite way, attempting to water down legislation, such as the EU AI Act, that are designed to improve the internet for everyone. Good technology improves governance and social welfare. Big Tech? Does that term include Alphabet (Google) and Apple, where our only grandchild’s parents work?
At Re:publica’s “maker space”, Eleanor fiddled around with DIY solar-powered sensors that can be built using a Raspberry Pi computer and off-the-shelf components such as humidity sensors. Her partner, an engineer by training, investigated a microscope designed by the OpenFlexure project that was made from 3D printed materials. Microscopes are crucial for diagnosing infections but can cost serious money, making them inaccessible for the poor. This one is lightweight, costs next to nothing and is open source, meaning that anyone can reproduce the design by 3D printing parts and cobbling them together with shop-bought motors and circuit boards.
Even in a prosperous country like Norway, it is important to have components available for everyone to use. This applies to many different categories of devices, including OpenFlexure’s designs to give people everywhere access to low-source microscopy. Increasingly, open-source is a necessary solution, as is a right to repair, for proprietary products.
Drage asks: Why does technology create new problems for each one it solves? She mentions that farmer Chris Conder dug her own fibreoptic cables on her property in Lancashire, “to prove that ordinary people could do it … it wasn’t rocket science”. By demonstrating that fast internet could be connected with fibre-optic cable, a digger and the desire to just get on and do it, she spawned an organisation called B4RN, which promotes community fibre partnerships.
Here at Vangshylla, in Inderøy, Norway, property owners were required to dig their own trenches when we were connected to fibre-optic cables. Optical fibres are widely used for communication because they permit transmission over longer distances and at higher bandwidths (data transfer rates) than electrical cables. Signals travel along fibre with less loss and are immune to electromagnetic interference. Digging between the houses involved the use of machines and trained operators. Fusian splicing was done by professionals. All of the houses in the community had basic wiring installed to the outer wall of the house, although those who opted out from connection, did not have it installed inside. This was an attempt at future-proofing the community. For the first five years, those who opted in were required to use the services of a single provider. After that there was competition between several providers.
I took a pause from writing to walk down to the library to find a book written by William Kamkwamba (1987 – ) and Bryan Mealer (1974 – ), The Boy Who Harnessed the Wind (2009). The cover explains Kamkwamba’s situation in 2002. Malawi was stricken with famine. He set out to provide his family with electricity and water. He did this by building a primitive windmill, that provided electricity that powered, in addition to other things, a water pump. Of all of the courses I have taken in my life, I am most thankful for industrial arts, which taught me basic skills in electricity and electronics, woodworking and metalworking. Even the draughting was probably useful. Computer-aided design (CAD) emerged in 1959. The best known products emerged in the 1990s: SolidWorks (1995), SolidEdge (1996) and Autodesk Inventor (1999). Open Source/ free varieties, such as FreeCAD (2001) became available in the 2000s.
The most outstanding feature of frugal innovation is that it shows people that tech can be developed locally. Technological innovation is more about collaboration than it is about competition. The enemy of innovation is passive acceptance of technology. Instead one must ask what kind of world we want to live in? Then determine how to create it.
The following content originates with Andreas Spieß, Why AI Agents Replaced the Arduino IDE in My ESP32 Projects (Claude Code, Gemini CLI, Codex), published on YouTube 2026-01-11. In terms of pure micro-controllers, my favourite YouTube channel involves Andreas Spieß or Spiess, from Basel, Switzerland. His amateur radio callsign is: HB9BLA. In the above post, he stated: The transition from “coder” to “architect” is the biggest productivity jump I’ve experienced in 40 years of electronics.
A course in microprocessor (µp) technology was part of my basic computer science education. Except, while it might have been called that, it involved the use of micro-controllers: expensive and delicate but reusable devices. Ultraviolet light was used to erase content, and reset the device. One had to use protective glasses to prevent eye damage. The course involved programming traffic lights at an intersection, with increasing sophistication. In 2025, micro controllers are inexpensive, and less invasive methods can be used to erase content.
It is time to distinguish microprocessors from micro-controllers. A micro-controller integrates the CPU, memory (RAM/ROM), and input/output peripherals onto a single chip. They are best for: Embedded systems, remote controls, sensors, and battery-powered applications. Their key advantages include: Highly energy-efficient, cheaper overall board cost, fast boot-up times, and suitability for real-time hardware control. Their limitations include: Limited memory and processing speed; cannot run full, modern operating systems.
In contrast, a microprocessor only acts as the central processing unit (CPU). It requires external chips (RAM, ROM, and I/O devices) to function. They are best for: Computers, including desktop and laptop machines, smartphones, gaming consoles, and heavy computational or graphical tasks. Their key advantages include : High processing power (often GHz range), robust multitasking, and ability to support rich user interfaces. Their limitations include: Consume much more power, requires a stable external power supply, and needs a much more expensive, complex circuit board.
I regularly have to thank my daughter, Shelagh, for enlightening me about Arduino micro-controllers. She used them while a student at Emily Carr University of Art and Design, probably starting in 2006. I started using them in 2008, and stopped close to my retirement at the end of 2016. After I retired I used Raspberry Pi devices. I have asked Derek, my son in law, to buy a couple of Teensy micro controllers, so that I can experience them as well.
Then there is ESP 32. ESP32 is a family of low-cost, energy-efficient microcontrollers that integrate both Wi-Fi and Bluetooth capabilities. I have not used them yet. These are made by Espressif Systems, which has its head office in Shanghai, China.
Not included, level 0: Vibe coding = ask a chat program to solve a problem, then copy/ paste the result, test and hope. Vibe coding is a software development workflow where one describes the goals = vibe of an application in natural language, and an AI assistant generates the code. The human’s role is to guide the AI through an iterative loop of prompting, observing results, and refining. The term was coined by the Slovak-Canadian AI researcher Andrej Karpathy (1986 – ) in 2025.
The ten levels are: 1) Spec driven development: Write specifications and allow the AI to write the code. If something doesn’t work, update the specification. Yes, one still has to copy and paste the code, while the compiler’s messages have to be given to the AI system being used. One may have to ask the AI system to provide the entire code, if only part of it is given. This works, but the chat window becomes messy. 2) Functional Specification Document (FSD): Create a file with specifications. Suggestion – use Markdown. Markdown is a lightweight markup language for creating formatted text using a plain-text editor. It was created by American technologist John Gruber (1973 – ) in 2004. Markdown uses annotations for headings. AIs can read and write such texts. Unfortunately, humans are not fond of the results. To reformat the results for humans, Speiß recommends the use of Typora (about $17 for 3 computers), and refers to the reformatted result as a FSD. He always includes FSD in the file name. EtherPad stands out as a leading open-source alternative to Typora. It is especially useful when working on collaborative documents. 3) Agents: AI chatbots are examples of Large Language Models (LLMs). An agent is any mechanism used to connect with LLMs. Anthropic (Claude Code), Google (Gemini CLI) and OpenAI (Codex) have agents specifically designed for programming, with the names shown in parentheses. These can be used in terminals or integrated into Integrated Development Environments (IDEs). Programming agents are designed to work with folders rather than files. They can read, write and edit individual lines. According to Andreas, Claude Code works best for programming. Since my daughter Shelagh works for Google, I may use Gemini.
4) Git Hub. This offers protection against accidental file corruption, and provides history tracking. Type instructions into a text window, and the LLM responds. The real work is done behind the scenes, not in the chat window. 5) The Boss vs the worker. All agents ask for permission to execute commands, then, if it is given, execute them in the terminal. Some systems only operate in a sandbox, reducing potential damage. 6) Virtual Machine. Speiß grants full access to the agent on his server. This means it can do everything it wants, but is in an isolated environment. Human tasks = writing tasks in plain English, supervising progress, verifying results, discussing approaches with the agent. This is not free, it costs Speiß about US$ 20/ month. Speiß noes not use an IDE, only a terminal window
At this point only a basic summary will be given of the levels. 7) Many parallel projects. 8) Command Line Interface. 9) Datasheets. 10) Some technical things here, related especially to Arduinos and ESP32 micro controllers. The main takeaway is that programming agents can handle this. They can also write documentation.
Hallucinations: These are not an issue in embedded software development. Compilers do not tolerate hallucinations, because they are errors. Humans are still needed for testing. However, since time is not wasted coding, there is time available for testing.