Panolayer

Understand how your codebase fits together, build with the agents you know, and verify changes as code is written, so you catch regressions before they land.

[Jason Ederle @yepthatsjason]

@Panolayer's Function-level Verification engine, will be launching a Visual Studio Code Extension soon.

This will integrate our custom AI verification models into most of your favorite browsers: @Cursor, @code @WindsurfCurrent , @kirodotdev, @antigravity and many others!

Here's a sneak peak to how easy it is to get started.

[Alex Sidorenko @asidorenko_]

Writing code by hand might be less relevant but understanding what you've built is not

[Panolayer @Panolayer]

@asidorenko_ Yes! 100% and that gap is getting wider esp when you get a 5,000 line diff. We solve that with visual architecture -- a live map of your codebase parsed from the source with every function, caller, symbol, dependency, etc. You see how everything fits together and what a change affects. And we run code verification on agent save, so you know what broke and why before you merge its work into your branch. Devs need to know what ships.

[Callan @callanxai]

Andrej Karpathy:

"Prompting is going away.

Delete everything, keep Graph."

In 1 hour he explains how to build Graphs, and why it's the only thing that will be left standing at the end

LLMs → Prompts → Agents → Graphs

a prompt gets you one answer and forgets the job

an agent finishes the job and forgets what it learned

a graph saves every finished job and turns it into the starting point for the next one

watch it today

then save the full guide on Graphs below before everyone catches up ↓

[Panolayer @Panolayer]

Same model, 4.4% → 61.1%, and the gain came from the harness. The catch is that most teams won't hand-code one. In Panolayer you build it as a graph: steps, checks, and what feeds into what, with no orchestration code. Custom harnesses shouldn't need a research team.

[elvis @omarsar0]

Learn to build custom harnesses, folks!

Here is an impressive paper showing the impact of a good custom harness on long-context work.

GLM-5.3-Flash goes from 4.4% to 61.1% exact match on OfficeQA Pro V2 with the same model.

The authors report that the gain comes from the harness.

The model-agnostic harness name is LongHarness, built for corpora of 100M tokens and more.

Why it works:

Standard agent harnesses keep evidence and conclusions mixed into one growing chat history. Retrieval keeps expanding without closing the actual gap, and corrections stop reaching the answer.

LongHarness stores both as explicit state. Open conclusions decide what to retrieve next, and each new piece of evidence confirms or revises the conclusions that asked for it.

On the same benchmark, it beats Claude Code by 12.1 points on V1 and Codex by 18.9 points on V2.

https://t.co/kkLcbh6dyT

Chat with Paper: https://t.co/s0U2Gb8Pfa

[Jason Ederle @yepthatsjason]

This is the struggle with AI coding... it builds something new, but breaks things that worked previously along the way. 🤦‍♂️

In the video you see Panolayer's Verification Engine blocked 1 regression from entering my codebase, that would have broken an existing feature. And based on the changes we traced all connected code to re-validate 108 functions. This is the power of Panolayer.

Catch bugs BEFORE they merge, not AFTER and when it's 1 click away from merging in a PR. Humans decide when to merge, and can easily access the risk of the changes.

cc @cline, @Panolayer

[Panolayer @Panolayer]

Trace a bug down to the line. Panolayer builds a visual architecture parsed from your code, showing every group, component and function and how they connect. A red badge marks the one with a failing rule, so you can follow it down to the exact line. See how Panolayer turns your codebase into a visual architecture: https://t.co/EtZGfHjOwh

[Jason Ederle @yepthatsjason]

Single home for All Coding Agents!

It's a fully open marketplace, just a config file on github to publish into it from any repo using panolayer-plugin tag.

22+ agents and counting including: @kilocode, @cline, @opencode, @claudeai

What agents am I missing???

[Panolayer @Panolayer]

@lightsilver323 We launched today! 🚀
https://t.co/nZDV0xSX4k

[Jason Ederle @yepthatsjason]

🚀 Today we’re launching @Panolayer, an #IDE designed for agentic engineering.

With Panolayer you can build custom agents or bring the ones you already use. Easily navigate a codebase you didn’t write and catch regressions BEFORE the agents even commit.

With Panolayer you can:
• Run agents in containers that work in parallel
• Verify agent code changes in real time
• Visually navigate the code and trace bugs to the line
• Build custom graphs and ship them as agents
• Review code quickly with visual risk graphs
• Living documents in sync as the code changes

Panolayer runs on custom-trained models and algorithms with deterministic outcomes that deliver function-level verification of your entire codebase.

Building and training our verification engine from the ground up has been a particular focus and passion of mine. Big thanks to the @nvidia Inception Program!

I'd love your feedback, good and bad. How are you keeping your codebase stable while wrangling a bunch of agents?

https://t.co/mJZh6KSX4C

#AICoding #DevTools #AgenticEngineering #AIAgents

[Panolayer @Panolayer]

Today we introduce Panolayer, an #IDE designed ground up for agentic engineering. 🚀

With Panolayer developers can build with the agents they already know and use, understand the code they didn’t write, and know what changed and what broke BEFORE the agent commits. You can also:

• Run agents in containers and work in parallel
• Verify agent code changes in real time
• Visually navigate the code, trace bugs to the line
• Build custom graphs and ship them as agents
• Review code quickly with visual risk graphs
• Living documents in sync as the code changes

Panolayer runs on custom-trained models and algorithms with deterministic outcomes that deliver function-level verification of your entire codebase. A big thank you to the @nvidia Inception Program! We've been part of it all through stealth.

You can download Panolayer for macOS today: https://t.co/Gjf4Im13mq

#AI #developertools #AIEngineering