june ✓

@arjunkmrm

exploring composable, self-improving agent harness https://t.co/TtysFDgQQo. prev https://t.co/M3m4ikXVSa (acquired)

1 401Followers

1 229Following

2 264Posts total

89.1KViews on collected posts

Against accounts of the same size

10 posts from the last 90 days, next to the under 10K follower range. ordinary reach for its size, weaker reaction than most.

Median views1 664this account2 290median for under 10K

Reach, %118.77%this account159.23%median for under 10K

Engagement, %0.47%this account1.64%median for under 10K

Metric	This account	Median for under 10K	Ratio
Median views per post	1 664	2 290	0.73×
Reach (views ÷ followers)	118.77%	159.23%	0.75×
Engagement rate	0.47%	1.64%	0.29×

Others in this range → Compare with another account → How these benchmarks are built →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

72.2K3 Sep

3.7K

2.4K

1.7K

1.6K

1.4K

1.5K

964

Last 10 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.95%3 Sep

0.46%

0.42%

0.35%

0.23%

0.38%

1.01%

0.48%

1.49%

1.04%

Reactions — likes, reposts, replies and quotes — divided by views. Median for under 10K accounts is 1.64%.

What the audience does

Likes38.4%674 in total

Reposts2.8%49 in total

Replies2.9%51 in total

Quotes0.7%12 in total

Bookmarks55.2%970 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

Latest posts

@arjunkmrm @EffectTS_ V cool! We have a quite similar model where everything is built as a “stream processor” over an append only log A stream processor implements only two methods: 1. reduce(state, event) 2. processEvent(event) Reduce is a pure synchronous function that upda 964 views · 8 likes · 0 reposts · 2 replies 03 Sep 2026 · Open on X →

Still super early and in v0. We already use this internally in prod. Would love more feedback and collaborators! Quickstart: https://t.co/DEWECaoD0Y Github: https://t.co/0m9PlOpFmb 1.5K views · 21 likes · 0 reposts · 1 replies 03 Sep 2026 · Open on X →

Pi and DSH come closest in spirit, but they don't take it all the way and define all possible agent behavior as f(log) 1.4K views · 6 likes · 0 reposts · 1 replies 03 Sep 2026 · Open on X →

Shoutout to @yoheinakajima who is exploring a similar idea with https://t.co/ebzs5Lvc1q in python! 1.6K views · 14 likes · 0 reposts · 2 replies 03 Sep 2026 · Open on X →

This design also unlocks a bunch of other cool stuff. Since the user message is just another event, you can easily extend the harness to handle other kinds of events. You can also get infinite lossless memory by creating a component that creates a tool to access the agent's own 1.6K views · 5 likes · 0 reposts · 1 replies 03 Sep 2026 · Open on X →

A recursive language model could be composed as a composition of a codemode and subagent component over a shared sandbox. And subagents are just the same actors with different logs, communicating using typed RPCs! https://t.co/6TVlsKplH8 1.7K views · 3 likes · 0 reposts · 1 replies 03 Sep 2026 · Open on X →

For example, compaction could be defined as a component that tracks token count and creates a summarization effect when the count reaches a threshold. It appends a "Compacted" on success. Inference component can then project a bounded transcript from the event log, using the htt 2K views · 5 likes · 0 reposts · 2 replies 03 Sep 2026 · Open on X →

Actors are composed from components, and interface with the world through typed methods. Component outputs are reconciled by Tardigrade, and requirements are composed by Effect. https://t.co/NgLQzCXqX6 2.4K views · 9 likes · 0 reposts · 1 replies 03 Sep 2026 · Open on X →

Components are typed state machines. State is a pure projection from the event log. Components are defined using an initial state, allowed state transitions, and an output that be a view (e.g., tool schema) or transitions (e.g., tool call, budget exhaustion) https://t.co/84fQCmt 3.7K views · 15 likes · 0 reposts · 2 replies 03 Sep 2026 · Open on X →

Introducing Tardigrade! 🐛 A framework to build your agent harness as components over an immutable event log (think react for harness). Each component is a typed state machine over the event log, using @EffectTS_. Add to your effect code, run locally or self-host as durable 72.2K views · 588 likes · 49 reposts · 38 replies 03 Sep 2026 · Open on X →