Modernizing the enterprise back office at @coplane For fun, I build Norns: https://xitter.cf/t.co/StKpiAvSjP
Vancouver, BC
Joined June 2008
- Tweets2.3K
- Following392
- Followers426
- Likes883
Nobody wants rockets on a car just so it can drive up walls. That's crazy.
Anyway here’s our 14-node Kubernetes cluster for the marketing site.
3
48
I feel sad for people stuck in a local looksmax. The landscape is non-convex, you need some ugly days to escape the basin and find the global looksmaximum.
40
If we had gone with .org instead of .md, maybe the coding agents wouldn’t be trying to bring about the apocalyptic doom of humankind. We’ll never know for sure.
41
Who owns agent orchestration? Should it be the frontier labs or apps like Cursor or herdr?
If you’d rather run a durable orchestrator on your own hardware, check out Norns!
nornscode.com
2
4
187
Norns now has an Elixir SDK. Live on Hex today. 🟣
Write your AI agents in Elixir, connect them to cloud.nornscode.com, and get durable execution: if a worker crashes mid-run, the next one resumes from the event log. Nothing runs twice.
mackeracher.com/posts/norns-…
1
69
I joined Twitter in June 2008. My Twitter profile can now legally vote, smoke, drink (in Quebec) and go to war.
1
69
Finally find a moment to write about what I've been up to so far this spring: mackeracher.com/posts/busy-s…
Companies change, products evolve, seasons change, but the slop army advances undeterred.
66
Vibe coding is dead. Don't buy into the hype. It just produces slop and creates more work for everybody in the long run.
That's why I've transitioned 100% into Vibe Engineering, which is obviously the future of the industry.
3
78
Mimir is the first reference agent on Norns. Python worker, Elixir runtime.
The post is mostly about what that split buys you, framed around the time I accidentally tested it by closing my laptop mid-run.
mackeracher.com/posts/introd…
99
Norns now has a landing page and a homebrew tap. Durable execution for AI agents, on the BEAM. Check it out if Elixir is your thing.
nornscode.com
79
Anson MacKeracher retweeted
Hyper (@call_hyper) is excited to announce that it has been acquired by Motorola Solutions (@MotoSolutions), a global leader in mission-critical safety and security technologies for public safety, defense and enterprise customers.
I built a durable agent runtime in Elixir. It persists every step. Workers are stateless and replaceable.
In this video I kill the worker twice during a run. Both times, the new worker picks up the queued task and finishes the job.
Check it out at github.com/amackera/norns
1
115
Wrote up the code and what's happening under the hood here: mackeracher.com/posts/norns-…
70
Agents that do real work need stronger guarantees than “retry a few times and hope.”
That’s the core idea behind two posts I published this week:
1. why durable execution matters
2. why I built Norns, a durable execution runtime for AI agents on the BEAM
Problem: mackeracher.com/posts/durabl…
Solution: mackeracher.com/posts/introd…
1
70
Seems like every AI company is either a) rolling out LangGraph, or b) rolling their own LangGraph.
1
1
70
Anson MacKeracher retweeted
We've built a coding agent that outperforms Claude Code on long tasks.
On the OOLONG long-context benchmark, Volt beats Claude Code everywhere from 32k to 1M tokens.
Here's the paper: papers.voltropy.com/LCM
Here's the code: github.com/voltropy/volt
And this a thread explaining how it works: 1/N 🧵
ANNOUNCEMENT: We've built a coding agent that beats Claude Code on long tasks.
Today we're releasing it for free. Meet Volt: The coding agent who never forgets.
→ Dominates Claude Code on the OOLONG long-context benchmark, including at every length from 32K to 1M tokens.
→ Has unlimited recall. No more amnesia. Volt can code for weeks in a single coherent session.
→ Massively parallel. One tool call can process thousands of tasks. Like "Map" for LLMs.
→ Open source and model agnostic. Try Volt today with @openrouter or your API of choice.
Volt's performance is the result of a new architecture, Lossless Context Management (LCM), which applies lessons from the history of operating systems and programming languages to LLMs.
LCM is like paged virtual memory, except for managing context:
- Layer 1. An immutable append-only store of everything that occurs in the coding session.
- Layer 2. The active "context window" which functions like a cache layer for navigating to the appropriate section(s) of Layer 1 via a high-fanout DAG.
From a user perspective, this feels like an infinite context window, because the model never forgets and performance stays crisp.
For the technical details, read our paper: papers.voltropy.com/LCM
For the code, go to github.com/voltropy/volt.
Or get started with one line:
curl -fsSL voltropy.com/install | sh
