January 2026

I entered the new year holding an inconsolable, shrieking baby while London set off an armageddon of fireworks around us. So goes parenthood. The baby is fine, just congested and teething. I am as “fine” as anyone can be after months of chronic sickness, broken sleep, and parental troubleshooting. I am very tired and full of stoic perspective, but still savouring the baby babble sounds, tiny fingers on my face, and three-teeth grins. I’m certain I’ll soon yearn for these early morning hours, curled up with a tiny, snoring infant on my chest.

Parenthood is a predictable source of exhaustion. But there’s a second, far less expected source in my life right now. And it doesn’t come with a cornucopia of adorable noises to take the edge off.

Agents. AI agents are all I can see, read, build, and think about. Coding agents. Research agents. Planning agents. Sub-agents. Multi-agent swarms. Orchestrator agents. Agentic memory. Agentic context management.

This immersion is almost entirely voluntary and specific to my situation. I started a new job at Github Next at the beginning of October; a team tasked with researching and building the next generation of tools for software developers. Which at this point in history unquestionably means agents.

It’s hard to think of historical parallels where a field changed this rapidly in such an unrelenting and distributed way. Even Andrej Karpathy feels behind

I am not trying to add to the hype and FOMO here. Only to be honest about what it feels like inside my particular information bubble. I am becoming a product of my X feed, which is unintentionally finely tuned to show an infinite stream of developer-flavoured AI panic anxiety that looks something like this:

A geometric pattern avatar

Look I just one-shotted a live data map of all UK trains

A geometric pattern avatar

Orchestrate forty multi-agent swarms concurrently in this new terminal tool

A geometric pattern avatar

SaaS platforms are DOA in 2026 – we’ll all prompt our own replacement

A geometric pattern avatar

Any developer who doesn’t understand context graphs will be destitute in a year

A geometric pattern avatar

MCP is dead, we’re all on skills now

A geometric pattern avatar

If you aren’t using OpenCode you’re NGMI

A geometric pattern avatar

You must read this repo of hundreds of best-in-class cross-platform agent skills

A geometric pattern avatar

Your broken RAG system is so 2024

A geometric pattern avatar

Look I just one-shotted an interactive simulation of a J series car engine

A geometric pattern avatar

Look I vibecoded a live radio app in an hour

A geometric pattern avatar

If you aren’t using memory beads you’re NGMI

A geometric pattern avatar

Anyone who hasn’t tried the latest Gemini model is living in a different reality

You might suggest that I spend less time on X, but I’m not inclined to look away just as the train gets up to full speed. Sure it’s a distorted reality, but it points to real ground truth: even if progress on language models slows this year, we are still far behind in using what already exists to reshape software design and engineering.

To be clear, I am tired, but thrilled by the capabilities overhang. No one has the full context of what is happening around us. Pick any piece of it to work on in earnest and you’ll find bushels of low hanging fruit.

I am not a resolutions person, but it’s hard to enter a new year without stopping to take stock and strategise a bit. My policy for the first year of my kid’s life is that I get a free pass at everything; eating too many chocolate Hobnobs? Free pass. Not reading enough books? Free pass. Haven’t cleared out that pile of crap in the hallway? Free pass. This excuses me from most new-years-shaped personal improvement goals.

But the one thing I’ve lost over the last nine months that I urgently need to find again is my belief that anything I write matters. It’s been hard to know what to say with a landscape changing this fast. It’s hard to gather my thoughts in a resource depleted state. It’s hard to believe my opinions have any legitimacy compared to the people working inside the foundation labs, while I scramble together information in between 3am feeds and nursery runs. I’ve lost a little of my confidence as a researcher and contributor to The Discourse. My intention for this year is to take my own advice and pick some low hanging fruit.