"20% task completion in 2025. 77.3% in 2026."

An AI agent is a software system that can take a sequence of actions using tools, without a human directing every step. Everything else is implementation detail.

In practice, it means an AI model can browse the web, write to a file, call an API, send a message, schedule something, and then do the next thing — completing a multi-step workflow the way a person would, but faster and without needing to be told what to do at each stage. The model has a goal. It has tools. It pursues the first using the second.

The infrastructure shift that made this real is specific. In late 2024, Anthropic released the Model Context Protocol — a standard that defines how AI models connect to external tools. Within eighteen months, 97 million installs. Every major AI provider ships compatible tooling. What began as an experimental protocol is now infrastructure — the same way HTTP went from a proposal to the thing the web runs on.

Google followed with Agent2Agent, addressing how agents communicate with each other — not just how a single agent uses tools, but how systems of agents coordinate. The building blocks for multi-agent workflows are in place.

The pattern in well-designed agent systems is consistent: the agent handles the volume, the human handles the judgment. A system designed for human oversight is a system that can be corrected, audited, and improved. That is what you want from any tool you depend on.

For builders: safety patterns are worth building in from the start. Read-only mode while developing. Approval gates on anything that writes, sends, or executes. Logging everything so you can review what happened. Standard practices in any competent engineering environment.

The original BMO is a character from a cartoon — a small, cheerful game console with a face, opinions, and a tendency to talk to itself when no one is watching. The idea of building one, a real one, has existed in maker communities for years. The difference is that the AI needed to make it actually feel alive is sitting on your desk, accessible, local, and free.

What you're building: A compact, conversational AI companion that runs entirely on your local network. It wakes on voice command, speaks back, maintains a persistent memory of your conversations, and can be housed in any enclosure you choose. No cloud dependency. No subscription. Your data stays on your device.

The software stack: Raspberry Pi OS Lite (64-bit), no desktop environment. Ollama for the local LLM — llama3.2:3b fits in 8GB with fast inference. OpenWakeWord for wake-word detection, feeding into Whisper.cpp for speech-to-text. Piper TTS for voice output. A lightweight SQLite database stores conversation history — a simple context injection script pulls the last 10 exchanges into each new prompt. This is what makes it feel like it knows you.

Assembly: Flash Raspberry Pi OS Lite to microSD. Install Ollama. Pull your model. Install ReSpeaker HAT drivers. Wire the speaker. Install OpenWakeWord, Whisper.cpp, Piper. Write the main loop: wake word → record → transcribe → Ollama query with memory context → Piper → speaker. Add a Pygame display script as a systemd service. Print or build your enclosure. Name it.

The first time it calls you by your name — because you told it your name three days ago and it didn't forget — something shifts. Not because it's conscious. Because it's continuous. It remembers. That single property, running locally on hardware you own, assembled with your own hands, is quietly extraordinary.

There is a photograph circulating in architecture communities that stops people mid-scroll. It shows a room — spare, deliberate, unmistakably livable — with floor-to-ceiling windows that open not onto a skyline or a garden, but onto deep space. A ringed planet hangs in the upper right corner. The light source in the room is the cosmos. There is a single chair, facing outward.

The photograph is a rendering. But the instinct behind it is real, and it is driving a quiet movement in interior design. Call it celestial minimalism, or cosmic brutalism, or simply the long exhale of living with too many walls. Whatever the name, designers are increasingly building interior environments that treat the universe as the primary material. Space is not a theme applied to a room. It is the room's emotional logic. The architecture structures itself around the view.

What you want in a habitat in high orbit is exactly what you want in a room right now: silence, light, and the feeling that nothing is wasted. Architecture in these environments is not a theme; it is a survival logic that becomes an emotional logic.

The materials language is consistent across its practitioners: poured concrete, brushed steel, raw timber, glass. No texture fights for attention. No colour competes with the view. The room's palette is deliberate vacancy — the decorator's equivalent of silence.

What does it do to a person to spend time regularly in a room that situates them within the universe rather than apart from it? The designers working in this space are consistent in their answer: it scales things correctly. The problem you came home carrying seems different when you've sat for ten minutes across from a nebula.

You don't own a cassette player anymore.

You hold it up to the light and through the tape window you can see the spools are lopsided — most of whatever was on here has already played. Someone listened to most of it. Someone stopped.

You put it back in the drawer. You leave it for another three years.
It is still there.

An ocean estimated to contain twice as much liquid water as all of Earth's oceans combined.

On October 14, 2024, a rocket left Earth carrying nine scientific instruments, more than 2.6 million human names, a poem, and a question that has been building since the 1990s. The spacecraft is called Europa Clipper. It is the largest planetary mission NASA has ever built, and it is currently en route to Jupiter's moon Europa — a world roughly the size of Earth's Moon, but fundamentally different.

Beneath its cracked, ice-covered surface, Europa holds a vast, saltwater ocean estimated to contain twice as much liquid water as all of Earth's oceans combined. The evidence has been building since the Galileo spacecraft orbited Jupiter between 1995 and 2003. Europa Clipper exists to find out what is in it.

The mission will arrive at Jupiter in April 2030, having used a gravity assist from Mars in March 2025 and another from Earth in December 2026. It will not orbit Europa directly — the radiation from Jupiter's magnetosphere would destroy the spacecraft within months. Instead, it will orbit Jupiter in a wide ellipse, swooping to within 25 kilometres of Europa's surface on 49 separate flybys.

What those instruments are looking for is specific. Ice-penetrating radar will map the thickness of the ice shell and the shape of what lies beneath. Magnetometers will measure how Europa's interior responds to Jupiter's magnetic field. Spectrometers will analyse surface composition and hunt for organic compounds. There may also be plumes — if Europa Clipper encounters an active plume, it could fly through it, sampling ocean water delivered directly to the surface.

Life appeared on Earth approximately 3.5 billion years ago — quickly, in geological terms, once liquid water and basic chemistry were present. Europa's ocean has existed for an estimated 4 billion years. It has an energy source: tidal heating from Jupiter's gravitational pull likely produces hydrothermal vents on the ocean floor. It has water. It may have organic chemistry. It has had all of this, stably, for longer than complex life has existed on Earth.

If Europa can do it, the universe is full of oceans. And oceans, we are learning, tend to do things.

It will arrive in six years.

The Millennium Falcon is, structurally speaking, a nightmare. Its irregular, asymmetric hull would create differential aerodynamic loads during atmospheric entry requiring active compensation at all times. Its sub-light engines produce thrust vectors that don't pass through the craft's centre of mass. The cockpit, mounted to the side, would expose the pilot to lateral g-forces that standard inertial dampeners would struggle to manage.

None of this is a reason not to build it. It's a reason to respect what building it would actually require.

Propulsion: The Falcon's sub-light drives are described as ion engines — a technology that is real, currently used in deep-space probes including Dawn and Hayabusa2. Real ion engines produce efficient thrust at very low magnitudes. A craft the Falcon's size would require a megawatt-class fission system. NASA's DRACO programme is currently developing kilopower-class nuclear reactors. Megawatt-class is further away, but not theoretically impossible.

Hull: The current leading candidate for a craft designed to operate in atmosphere and space simultaneously is carbon-fibre-reinforced polymer with a metallic thermal protection system. Given that the Falcon is famously always breaking down, CFRP's limited repairability is a problem. The Falcon should probably be stainless steel.

Inertial dampeners: The forces implied in a Kessel Run scenario are orders of magnitude higher than anything a pressure suit handles. The fictional inertial dampener has no real-world equivalent, though research into active magnetic compensation for neural tissue during high-G events is ongoing.

85% of a Millennium Falcon, by the end of this century, is a reasonable engineering target. The 15% is left as an exercise for whoever figures out negative energy density.

South Africa added nearly 9 gigawatts of solar capacity in 2024. The private sector contributed 6.1 gigawatts of that. These are not government projects; they are thousands of individual installations, each one a separate decision about power.

A standard residential rooftop system runs between 5kW and 15kW. A commercial installation might be 50kW to 500kW. To reach 6.1GW of private solar, roughly 600,000 to 1.2 million individual systems were installed across the year. Each one needed: panels (monocrystalline, 400W–550W per panel), an inverter, mounting hardware rated for local wind loads, surge protection, and either grid-tie approval or a hybrid inverter with battery storage.

The grid-tie process varies by municipality. Some require an application, a certified installer’s sign-off, and a bi-directional meter. Others have no formal process at all. The difference between a system that feeds back to the grid and one that islands itself during outages is a single setting in the inverter.

On the engineering side, rapid growth has created tensions. Inverter harmonics can backfeed into neighbourhood grids. Early adopters who installed panels five years ago are seeing degradation curves 3–8% below manufacturer projections. The equipment is mostly imported, but local frame assembly is growing. Solar PV generated 5.2 TWh in 2024 — roughly what a 600MW coal plant produces in a year.

The ultimate interface for the Federated Brain. A split mechanical keyboard that knows when you're typing to an agent and when you're typing to a human. The OLED display provides a real-time heartbeat of your local inference engine.

Five tools. One person. The same output as a team of twelve six months ago.

A feature film shot in Cape Town runs on a specific set of calculations. Crew rates are roughly 40% below Johannesburg and 60% below Los Angeles for equivalent roles, but the pool is smaller — a good gaffer might be booked six months out. Equipment rental houses stock Arri and Sony bodies, but not every model. If the script calls for a specific lens, it may need to come from London or Johannesburg, adding a week to prep and R15,000 to the budget for insurance and courier.

Post-production is where geography matters most. Cape Town has several colour grading houses on Baselight systems. Sound design and mixing can be done locally. But VFX-heavy work still routes through London or Mumbai, because local VFX houses are built for commercial spots at 30-second increments, not feature-length at 2,000 shots.

The streaming platforms changed the math. When a show is greenlit for Netflix, the budget is set in US dollars and managed through a local service producer. A budget that covers a four-episode limited series in London covers eight to ten episodes in Cape Town, with the same camera bodies and a crew that has been working on international productions for twenty years.

The real constraint is not talent or equipment. Cape Town has roughly four sound stages large enough for a Netflix series, and they are almost never empty. If a production does not book studio space twelve months ahead, it films on location or it waits.

When the brush moves, the desk resists. Haptic feedback in the creative process is the bridge between the abstraction of AI and the reality of the hand. The Monolith maps latent space parameters to physical resistance.

This week, in a suburb of Johannesburg, a 29-year-old woman drove a truck to a house that had never had reliable electricity and installed the hardware to change that. She did this without a pitch deck. Without investors. Without a press release.

In a shipping container in Cape Town, a team of four is building a soil sensor that can tell a small-scale farmer exactly which three nutrients their specific plot is currently missing. They are not building this for agribusiness. They are building it because they grew up near farms.

On a Raspberry Pi, somewhere in a bedroom, a person is running a local language model for the first time and discovering that the AI they've been reading about in headlines is actually just software, sitting on their desk, waiting to be useful.

These things are happening now, in parallel, unreported, unglamorous, consequential.

SOL.mag exists to point at them. Not to explain what they mean. Not to tell you how to feel about them. Not to package them into a trend with a name and a forecast. Just to find the things that are moving and show them to the people who are already looking in that direction.

You are one of those people. That's why you're here.