Skip to main content

Field guide

Reviewed August 2026

Current technology, chosen for the job.

This is not a leaderboard and nothing here is ranked. It is the short list of what gets used, grouped by the job it does, so you can decide and get back to work.

Write the code

Use an agent that can stay in the codebase, use tools, and leave a result you can review.

Devin

Cognition

Use for autonomous multi-step implementation, visual debugging, and working a real task through to a reviewable result.

Codex

OpenAI

Use for codebase work, implementation, review, and tool-driven engineering from terminal or IDE.

Claude Code

Anthropic

Use for deep repository understanding, careful refactors, and long-context implementation work.

Think it through

Pick the model to fit the task, then verify anything that matters against a primary source.

ChatGPT

OpenAI

Use for general reasoning, drafting, multimodal work, and broad problem-solving.

Claude

Anthropic

Use for long documents, structured thinking, and careful language work.

Perplexity

Perplexity

Use when every important answer needs visible sources and a fast web-research loop.

Make the images

Treat image models as art direction tools. They do not replace the judgement about what the picture is for.

GPT Image 2

OpenAI

Use for high-fidelity art direction, compositing, edits, and production-ready visual concepts.

Midjourney

Midjourney

Use for fast visual exploration when you need to test a direction before producing it.

Adobe Firefly

Adobe

Use when the output must continue into an Adobe production workflow.

Make the video

Choose on audio, motion control, take length, and how stable the production path is under load.

Veo 3.1

Google DeepMind

High-quality video with native audio and real creative controls.

Runway

Runway

Creative production and branded video workflows.

Kling

Kuaishou

Short-form motion, social content, and multi-shot work.

Luma Dream Machine

Luma

Fast visual exploration and image-to-video passes.

Run the agents

A useful agent needs memory, a real work surface, and a hard boundary on what it can spend.

Hermes

Nous Research

JoePro’s first-choice multimodal agent harness for durable memory, tools, and long-running work across real workflows.

Daytona

Daytona

Disposable development environments for isolated tool and browser work.

OpenClaw

Open source

An open-source agent runtime for local and cloud workflows.

Built here

These are mine. They are on the list because they do the job, and flagged so you can weigh that for yourself.

clawd.run

clawd.run

Live Cloud runtimes for Hermes and OpenClaw agents with persistent memory and a real computer.

Carapace

carapace-xi.vercel.app

Live A deterministic memory-integrity firewall and trust ledger for memory-first agents.