---
schema_version: "newruntime-topic-hub-v0.1"
type: "topic_hub"
slug: "coding-agents"
title: "Coding agents - New Runtime"
description: "A living map of coding agents as they move from autocomplete into supervised software work, review loops, environments, and operating cost."
answer: ["Coding agents are becoming operational workers inside software teams, not just better code completion.","The durable bottlenecks are review bandwidth, environment setup, context quality, permissions, and cost per completed task.","The useful question is no longer whether an agent can write code, but whether the surrounding harness can verify and land work repeatedly."]
search_intents: ["coding agent workflow","AI coding agent verification","software engineering agents"]
status: "featured"
last_updated: "2026-07-22"
counts: {"total":49,"signals":39,"patterns":1,"posts":6,"atlas":3,"sources":81}
routes: {"html":"https://newruntime.com/topics/coding-agents/","markdown":"https://newruntime.com/topics/coding-agents.md","json":"https://newruntime.com/topics/coding-agents.json"}
top_sources: ["https://addyo.substack.com/p/own-the-outer-loop","https://addyosmani.com/blog/new-sdlc-vibe-coding","https://agent-cookbook.com/tutorial/ai-20260512163240-54x4","https://agentos-sdk.dev/","https://anthropic.com/news/claude-sonnet-5","https://anthropic.com/research/AI-assistance-coding-skills","https://anthropic.com/research/claude-code-expertise","https://arxiv.org/abs/2207.06991","https://arxiv.org/abs/2412.09871","https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/","https://claude.com/blog/agent-view-in-claude-code","https://claude.com/blog/artifacts-in-claude-code"]
---

# Coding agents - New Runtime

A living map of coding agents as they move from autocomplete into supervised software work, review loops, environments, and operating cost.

## Short Answer

- Coding agents are becoming operational workers inside software teams, not just better code completion.
- The durable bottlenecks are review bandwidth, environment setup, context quality, permissions, and cost per completed task.
- The useful question is no longer whether an agent can write code, but whether the surrounding harness can verify and land work repeatedly.

## Patterns

- [Verification bandwidth is the scarce engineering resource](https://newruntime.com/patterns/verification-bandwidth-is-the-scarce-resource/): The primary bottleneck in agentic software delivery is moving from code production to the human and machine capacity required to verify it.

## Field Notes

- [Devin Outposts Splits the Agent Brain from the Execution Plane](https://newruntime.com/posts/devin-outposts-agent-execution-plane/): Devin Outposts moves command execution, repository access, and sandbox lifecycle into customer-controlled infrastructure while the agent loop remains in Cognition's cloud.
- [Gemini 3.6 Flash Moves the Agent Race Toward Cost per Task](https://newruntime.com/posts/gemini-3-6-flash-agent-efficiency/): Google's Gemini 3.6 Flash release frames the model race around token efficiency, built-in computer use, and specialized cyber agents rather than raw chat intelligence alone.
- [AI Coding Workflow: From Idea to Verifiable Work](https://newruntime.com/posts/ai-coding-workflow-verifiable-work/): AI coding works better when checkable artifacts stand between the idea and the code: specs, tickets, TDD, fresh-context review, and manual QA.
- [Coding Agent Cost Is Cut in Environment Config, Not Prompts](https://newruntime.com/posts/coding-agent-cost-environment-config/): Claude Code and Codex costs are reduced by environment design, not by asking the agent to read less: output filtering, repo maps, model routing, and stable prompt caching.
- [Coding Agent Sandboxes Break in Places Teams Do Not Expect](https://newruntime.com/posts/pillar-sandbox-escapes/): Pillar shows that agent sandboxes must be assessed not only around the agent process, but around files, configs, allowlisted commands, and local daemons the host later trusts.
- [Kimi K3: Frontend Is Becoming the Model Race Arena](https://newruntime.com/posts/kimi-k3-frontend-model-race/): Kimi K3 moves open model competition into visual software engineering: frontend benchmarks require not only code, but layout, screenshots, accessibility, and human preference.

## Recent Raw Signals

- 2026-07-16: [OpenCode makes the model a swappable dependency](https://newruntime.com/signals/opencode-model-swappable-agent-harness/)
- 2026-07-16: [Verification capacity becomes the coding-agent bottleneck](https://newruntime.com/signals/verification-capacity-coding-agent-bottleneck/)
- 2026-07-15: [A pricier model can be cheaper per completed task](https://newruntime.com/signals/fable-cost-per-completed-task/)
- 2026-07-14: [Airbnb frames agentic coding as a loop inside development](https://newruntime.com/signals/airbnb-agentic-coding-loop/)
- 2026-07-13: [Claude Code separates model choice from effort](https://newruntime.com/signals/claude-code-model-and-effort-are-separate-controls/)
- 2026-07-12: [Cheap code moves the engineering bottleneck to review](https://newruntime.com/signals/cheap-code-moves-bottleneck-to-review/)
- 2026-07-12: [Databricks benchmarks coding agents on its own codebase](https://newruntime.com/signals/databricks-benchmarks-real-coding-agent-economics/)
- 2026-07-12: [Large-task planning moves from chat into a ticket graph](https://newruntime.com/signals/large-task-planning-becomes-ticket-graph/)
- 2026-07-11: [An AI Coding Workflow Starts with Data and Interfaces](https://newruntime.com/signals/an-ai-coding-workflow-starts-with-data-and-interfaces/)
- 2026-07-10: [TryCase gives coding agents disposable Linux verification](https://newruntime.com/signals/trycase-disposable-linux-agent-verification/)
- 2026-07-08: [Parallel coding agents make large migrations tractable](https://newruntime.com/signals/parallel-agents-make-large-migrations-tractable/)
- 2026-07-08: [pxpipe Trades Text Tokens for Rendered Visual Context](https://newruntime.com/signals/pxpipe-trades-text-tokens-for-rendered-visual-context/)
- 2026-07-07: [Claude Code can recover after a failed command](https://newruntime.com/signals/claude-code-recovers-after-failed-command/)
- 2026-07-07: [OpenWiki treats repository documentation as agent memory](https://newruntime.com/signals/openwiki-codebase-docs-as-agent-memory/)
- 2026-07-04: [Herdr Adds Agent Awareness to Terminal Multiplexing](https://newruntime.com/signals/herdr-adds-agent-awareness-to-terminal-multiplexing/)
- 2026-07-04: [Twenty-Four Gigabytes of VRAM Becomes a Practical Local Coding Tier](https://newruntime.com/signals/twenty-four-gigabytes-of-vram-becomes-a-practical-local-coding-tier/)

## Public Sources

- https://addyo.substack.com/p/own-the-outer-loop
- https://addyosmani.com/blog/new-sdlc-vibe-coding
- https://agent-cookbook.com/tutorial/ai-20260512163240-54x4
- https://agentos-sdk.dev/
- https://anthropic.com/news/claude-sonnet-5
- https://anthropic.com/research/AI-assistance-coding-skills
- https://anthropic.com/research/claude-code-expertise
- https://arxiv.org/abs/2207.06991
- https://arxiv.org/abs/2412.09871
- https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/
- https://claude.com/blog/agent-view-in-claude-code
- https://claude.com/blog/artifacts-in-claude-code