AI Agents & Automation: RAG, Workflows & Guides

If you use GPT-6 Sol in Codex, switching to GPT-6.1 Sol is fine. Released on September 29, 2026, it is recommended for complex coding and long-running work, with the same context limit and standard API prices, half-price cached input, and a knowledge cutoff 10 days newer. The guide covers switching steps in Codex, the difference between new and existing chats, Plus and Pro usage limits, and how to compare both at Medium effort before you decide.

I had Claude Code Projects (public beta) build an entire website for looking up how much memory local LLMs need. From creation to the last merge took about 20 hours: 10 threads, 11 PRs, about 23,000 lines, and about 190 million tokens (97.7% of them cache reads). Three tasks ran in parallel overnight, the threads coordinated with each other through the coordinator, and once I switched to instructions that handed things over, it decided the next tasks by itself. On the other hand, nobody built the logic that decides "can this be published?" at the boundary between the data collection and page threads, so production showed zero models. Real data also exposed bugs that sample-data tests never showed, and deploying to a server reachable only over SSH hit a wall. I record it all from measurements: where I tripped getting started, the difference between approving every step and handing it over, how the setup credit ran down, and what kinds of work it suits and does not suit.

Claude Code cloud sessions came out of research preview and became generally available on September 23, 2026. Your GitHub repository is cloned into a virtual machine in Anthropic's cloud, where Claude fixes the code and pushes a branch, so the work doesn't stop when you close your laptop. Usage comes out of the same limits as your regular plan, with no separate charge for the VM. On top of that, existing Pro subscribers are getting a one-time $100 credit and Max subscribers $250: claim it by October 7 and it is spent first until November 5, 07:59 UTC (Projects and Routines are excluded). This article uses the official documentation and posts from the official account to cover how cloud sessions differ from local Claude Code and Remote Control (which settings come along and which don't, permission modes, the four network access levels and where to keep secrets), how to start from the browser and the terminal, how to claim the credit and the caveats in the full offer terms, and which jobs they suit and which they don't.

OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026 (US time). Sol is priced at 1/5 of Astra ($2 input / $10 output) and Luna at 1/100 ($0.10 input / $0.50 output), and both are available in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu, and in the API. Free and Go get Luna only, in the desktop app. Neither model was in regular Chat at launch, but since October 7 GPT-6 in Chat runs on Sol for Plus, Pro, Business and Enterprise, and since October 8 on Luna for Free and Go. GPT-5.6 remains available during the transition period. Written for people who currently use Astra at low (shown as Light in the app), this article separates its sources by publisher: the official use-case guidance (Sol at Medium, Luna at High, Astra from Light) and how it differs from the app's default preset (Sol Light); OpenAI's AutomationBench result where Sol xhigh beat Astra low; the values in Artificial Analysis's comparisons (42% for Astra low vs 19% for Sol medium on Terminal-Bench 4.0, but 59% vs 58% on business workflows); and AI Arte's small-scale test (every setting passed the short tasks, which is no basis for a ranking). It also explains why unit-price ratios don't carry over directly to the total cost of a job or to how much a subscription covers, and where the decision turns: Sol medium for everyday work, Sol low or Luna high for short, easy-to-verify jobs, and Astra for hard design work and important reviews.

Jev is TypeSafe's AI model that returns a choice, score, or Yes/No probability instead of generating text. This guide explains how Choice, Score, and Noul differ, how to read confidence, and how to write an API request that routes inquiries. It also covers documented weaknesses such as calculations, dates, and leading prompts, plus pricing, input limits, and how to evaluate Jev with non-English text, so you can judge whether it fits your work.

Projects in Claude Code has been rebuilt. Until now a project was a folder holding conversations and reference material; the new Projects is a single conversation. You write what you need, Claude splits it into threads, the threads run in parallel in the cloud, and each one opens a pull request and reports back when it finishes. Closing your laptop does not stop them. Three things are worth checking before you jump in, though: the accounts that can use it are still limited (a Pro and Max public beta, going out first to accounts with no existing projects), github.com and the Claude GitHub App are requirements in practice, and the rate at which it eats your usage limit is nothing like a single session. This article works through how to tell whether the rollout has reached you, what a thread starts with (including the trap where permission rules and hooks stop applying once a project holds more than one repository), where the token cost comes from including the default of Opus at high effort, and how to choose between the five ways to work in parallel: subagents, agent view, agent teams, dynamic workflows and Projects, all from the official documentation and blog.

A newer model is smarter but more expensive, or so the thinking usually goes. Yet comparing GPT-6 Astra with GPT-5.6 Sol on unit price alone settles nothing, because Astra finishes the same job on fewer tokens. In measurements by the independent benchmarking firm Artificial Analysis, running Astra at its weakest setting, low, costs almost exactly what Sol costs at high for one task ($0.82 against $0.81) while scoring higher (46 against 42), on one third the output tokens and a quarter of the wait before the reply starts. Working from the figures as of September 18, 2026, this article covers cost, tokens and speed at each level of reasoning effort, the breakdown that explains why a 2.5x unit price still lands on the same total, what the numbers look like once Sol's promotional price ends, the way the gap in price narrows on coding agent work, and the steps for measuring all of it on your own work.

OpenAI released GPT-6 Astra on September 3, 2026. The API is generally available with no gating, the model ID is gpt-6-astra, the context window is 1,050,000 tokens, pricing is $10 in / $50 out per million tokens, and the knowledge cutoff is April 30, 2026. But "it launched" and "your plan can use it" are two different things: on ChatGPT Plus it does not appear in the regular chat interface, and the way in is ChatGPT Work and Codex. This article covers availability by plan, how to think about pricing when a 2x unit price can still be cheaper per task, the benchmarks OpenAI actually named, the cybersecurity capability that reached "Critical" under the Preparedness Framework for the first time and the line OpenAI drew around it, the four things that break when you migrate (sampling parameters removed, Responses API, no none reasoning effort, the renamed cache setting), and a comparison with Claude Opus 5, Fable 5.1, and Gemini 3.8 Flash — restricted throughout to what could be confirmed against primary sources. It also shows, with each publisher named, how the same DeepSWE v1.1 gives Claude Opus 5 73.7% in the OpenAI table but 68.8% in the Anthropic announcement.

Dispatch is the feature where you send an instruction from your phone and Claude carries the work out on your own computer (beta, Pro and Max). It does not run in the cloud; your real machine moves, and that single fact produces both the value and the danger. The official help says you can message Claude from your phone and have it work on your desktop computer, using the same connectors, plugins and file access you have already configured in Cowork, inside what Anthropic frames as one continuous conversation reachable from either device. Running it requires the PC to be awake with the desktop app open, and computer use is supported on macOS and Windows only, with no computer use on Linux. Mechanically it works down three tiers of priority: a connector if one is available, browser navigation if not, and direct screen interaction as a last resort, with screenshots taken along the way to understand the display. The assessment starts after that. The places it stops are designed in: computer use is off by default and enabled under Settings, General; permission is asked for each new application; permanently deleting a file requires explicit permission; and investment and trading platforms and cryptocurrency apps are off-limits by default. But there are places it does not stop. Individual actions inside an already-approved app are not confirmed with you, and the official wording is that Claude clicks, types, and navigates your screen directly, without the permission checks that gate other Cowork tools. The docs add that there is no sandbox between Claude and what is on your screen, and that actions taken in one app can impact other apps. The largest risk is prompt injection, which Anthropic describes in its own words: web content is a primary vector for prompt injection attacks, and a manipulated instruction, an unexpected command, or a phishing link opened in your browser could cascade into actions that are difficult or impossible to undo. Anthropic says it scans model activations to detect such behaviour, but that lowers the odds rather than removing the need for you to draw a line, and the guidance still says to switch to manual approval whenever a task touches sensitive files, accounts or sites. Anthropic names the boundary outright: do not give computer use permission access to sensitive apps such as banking, healthcare and government, and avoid financial accounts, legal documents, medical information and personal data. The article also covers the phone side. What leaks if you lose the handset is not data stored on it but the standing to instruct your PC, plus the contents of the continuing conversation — and the official Dispatch help does not document unpairing or lost-device handling, so the remedies come from the account side instead: terminating the individual session under Settings, Account, Active sessions, logging out of every session from claude.ai (which is not available in the mobile apps and therefore needs a web browser), or simply cutting the PC side by closing the desktop app or letting the machine sleep, which is in fact the fastest because Dispatch needs the PC awake and the app open. It closes by separating Dispatch from computer use as two distinct switches, distinguishing both from Claude Code's agent view (which the official docs also call dispatch), and drawing a practical line: start with work you can take back.

Claude Code's agent view, opened with claude agents, is the feature for starting independent background sessions one after another and managing them from a single screen. The official documentation calls the operation you perform there dispatch, which collides with the separate desktop-app feature of the same name, so the first job is telling them apart. The docs describe agent view as the feature that lets you dispatch and manage many Claude Code sessions from one screen, and it is a research preview requiring v2.1.139 or later. This article sticks to the mechanics and the safety model. The first surprise is that every prompt typed into the input box starts its own new session: type a second one and you get a second session beside the first, not an extra instruction added to it. Further instructions go through the peek panel, opened with Space, which shows the latest output or the question the session is waiting on rather than the whole transcript. The heart of the safety model is isolation by worktree. Before editing any file, a background session moves into an isolated git worktree under .claude/worktrees/, so parallel sessions read the same checkout but each writes to its own — reads shared, writes separated. Anything that would reach the main checkout is cut off by three checks: file edits through Edit, Write and NotebookEdit; command working directories that resolve to the main checkout or that cannot be verified to stay outside it; and attempts to redirect git through git -C, --git-dir, GIT_DIR, GIT_WORK_TREE or a cd placed before the git call. The call is deliberately made on the safe side, refusing what it cannot verify, and the same protection is inherited by every subagent the session spawns. It is not an OS-level wall, though: files outside the repository and the network are out of scope, and PowerShell commands get only the working-directory check. Permissions are not chosen at dispatch time either; they depend on how the session starts (an agent view opened in a shell and claude --bg follow a normal launch's rules, one opened with ← uses the target directory's defaultMode or the originating session's mode, and /bg keeps the current mode), which means the looser your usual configuration, the more unattended loose-permission sessions you create at once. Three things then leak out of the isolation. Choosing "Yes, don't ask again" saves the rule into the main checkout's .claude/settings.local.json, so it applies in the main checkout and in every other worktree and survives the removal of the worktree it was made in. Deleting a session in agent view deletes the Claude-created worktree with it, so uncommitted work disappears — and Ctrl+X stops on the first press and deletes on the second. And .worktreeinclude copies gitignored files such as .env into every new worktree, multiplying your credentials by the number of sessions you dispatch. On top of that, quota drains in proportion to parallelism (ten agents use it roughly ten times as fast), and sessions run locally, surviving sleep but stopping when the machine shuts down. The article closes by placing agent view among the five official ways to parallelise, alongside subagents, agent teams, dynamic workflows and Projects, and gives a concrete routine for before, during and after a dispatch.

ChatGPT Work is the "AI agent for work" that OpenAI announced on July 9, 2026 alongside GPT-5.6. Unlike a regular chat that just answers, it gathers context from your connected apps and files to build finished deliverables — documents, spreadsheets, slides, and even web apps. It launched on GPT-5.6, then GPT-6 Astra (September 3) and GPT-6 Sol and Luna (September 22) were added, so paid plans can pick anything from the top-tier model to balanced and low-cost ones in Work to fit the task. Astra is limited on Plus and Business Standard, while Pro and Business Premium can spend their full allowance on it (OpenAI Help, as of September 25, 2026). Work differs from regular chat in its allowance (shared with Codex) and in which models you can pick; Codex offers the desktop app, CLI, IDE, and Cloud tasks on web and phone, while Work runs on web, phone, and desktop. OpenAI's wording on Free and Go is inconsistent. Built on OpenAI's own announcement, Help Center, and release notes, this article lays out what ChatGPT Work is, how it differs from regular ChatGPT and Codex, which models each plan can use, and which apps it connects to — with dates and confidence labels.

An in-depth comparison of GPT-5.6 Sol, OpenAI's flagship as of July 2026, and Google Gemini. Unlike the earlier battles against Claude, their strengths barely overlap: Sol dominates agentic and terminal coding (Terminal-Bench 2.1 88.8% vs 70.7%, SWE-bench Pro 64.6% vs 54.2%, both from OpenAI's evaluation table), while Gemini counters with native multimodality (voice and video) and a unit price about 40% of Sol's ($2/$12 up to 200K tokens vs $5/$30). There is also an important "timing trap": Google's true challenger, Gemini 3.5 Pro, was not yet released as of this writing (GA had been planned for mid-July 2026, and it was still unavailable on September 22), so the comparison target is Gemini 3.1 Pro (February 2026, still a preview in the API). This article lays out a spec cheat sheet, coding/reasoning/multimodal benchmarks, the multimodal gap that is Gemini's home turf, real cost (Gemini about 40% of Sol; on the GPT side Terra matches Gemini), a strengths-and-weaknesses map, and use-case-based selection, grounded in official announcements and independent benchmarks. For the successor, GPT-6 Astra (September 2026), see the note at the top.