AI Agents & Automation: RAG, Workflows & Guides

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 are inherited from that directory's defaultMode, or from a dispatched subagent's frontmatter permissionMode, 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 four official ways to parallelise, alongside subagents, agent teams and dynamic workflows, 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. Its brain is the new flagship GPT-5.6 Sol (built on Codex), and it can break a complex project into steps and keep working for hours (Thurrott). Inside the unified desktop app, three modes coexist — Work (deliverables), Codex (technical, shows details), and regular chat (conversation) — with Work positioned as the "business version" that hides Codex's technical detail (9to5Mac). The model depends on your plan: Free/Go default to Terra, while Plus/Pro/Business/Enterprise choose from Sol/Terra/Luna (as reported). Its strength is pulling in "your context" via connectors like Google Drive, SharePoint, and Slack, plus MCP — and for enterprise, data isn't used for training by default. Drawing on Axios, TechCrunch, 9to5Mac, Thurrott, and OpenAI, this article lays out what ChatGPT Work is, how it differs from regular ChatGPT and Codex, which plans can use it, and which apps it connects to — with confidence labels on what is not yet official.

An in-depth comparison of OpenAI's flagship GPT-5.6 Sol 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 68.5%, SWE-bench Pro 64.6% estimated vs 54.2%), while Gemini counters with native multimodality (voice and video), roughly half the price ($2.50/$15 vs $5/$30), and a lead on MMLU 92.6%, ARC-AGI-2 77.1%, and WebDev Arena. There is also an important "timing trap": Google's true challenger, Gemini 3.5 Pro, is not yet released as of this writing (GA planned for mid-July 2026 after a full architecture overhaul), so the fair comparison target today is the current flagship Gemini 3.1 Pro (February 2026). 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 2x cheaper than Sol, but Terra now undercuts Gemini), a strengths-and-weaknesses map, and use-case-based selection, grounded in official announcements and independent benchmarks.

Arriving July 9, just two and a half months after April 2026's GPT-5.5, GPT-5.6 is not merely a performance bump — the biggest change is the reorganization from a single flagship into a three-model lineup: Luna/Terra/Sol. What matters most to many GPT-5.5 users is that the mid-tier Terra delivers GPT-5.5-class quality at a far lower price: it launched at half price, and the July 30, 2026 price cut took it to 40% of GPT-5.5's unit price ($2/$12). For those wanting more performance there is the top-tier Sol (same $5/$30, with SWE-Bench Pro rising an estimated 58.6→64.6%); for those wanting lower costs there is Terra. This article lays out, based on official announcements and independent analysis, what the three-model shift means, a spec-at-a-glance table, Terra's price-to-performance, generational benchmark gains (same-metric SWE-Bench Pro +6pt, but TerminalBench differs in version 2.0→2.1 and much general reasoning is undisclosed), 5.6's new features (Programmatic Tool Calling, max effort, ChatGPT Work, GPT-Live, GitHub Copilot support), real cost (roughly $1,100→$440 for 100M input/20M output a month), and which model you should migrate to (cut the bill by 60% with Terra, then move only performance-critical work to Sol).

An in-depth comparison of OpenAI's flagship GPT-5.6 Sol (July 9) and Claude Fable 5 (June 9), which Anthropic positions as "the most powerful model it has ever made generally available." Where the Opus 4.8 matchup was a head-to-head in the same price tier, this one turns on a cost-versus-capability trade-off: "the half-price all-rounder Sol ($5/$30)" against "the twice-as-expensive but top-tier Fable 5 ($10/$50)." On production-grade coding's SWE-Bench Pro, Fable 5's 80.3% pulls more than 15 points ahead of Sol's 64.6% (estimated) — a gap wider than in the Opus 4.8 matchup. Fable 5 also self-drives for up to 12 continuous hours while focusing on millions of tokens, with Stripe finishing a 50-million-line Ruby migration in a single day as its home-turf "follow-through." Sol, meanwhile, leads on terminal operation (TerminalBench 2.1 88.8% vs Fable 86.0%), Agents' Last Exam (53.6 vs 40.5), and Coding Agent Index (80 vs 77.2), plus best value with half the price and +54% token efficiency. This article covers the spec cheat sheet, benchmark details, the "undisclosed-benchmark problem" of OpenAI withholding Sol's SWE-bench Pro, long-running autonomy, real cost (viewed per completed task), a strengths-and-weaknesses map, and how to choose by use case — all grounded in official and independent benchmarks.

An in-depth comparison of 2026's two AI-coding giants, Claude Opus 4.8 (May 28) and GPT-5.6's top-tier Sol (July 9). Their strengths are almost opposite: Sol leads in terminal operation and overall agentic capability (TerminalBench 2.1 88.8% vs Opus 78.9%, Agents' Last Exam 53.6, Coding Agent Index 80), while Opus 4.8 leads in production-grade coding, math, and long context (SWE-bench Pro 69.2% vs Sol 64.6%, USAMO 2026 96.7%, GraphWalks 1M 68.1%) and foregrounds honesty (overconfidence cut to one-tenth, 0% uncritical reporting of flawed results). OpenAI also leaves many of Sol's benchmarks undisclosed (SWE-bench Pro, GPQA, AIME, MMLU), so in coding's heartland the disclosed Opus has the edge. We cover the spec table, benchmark details, the undisclosed-benchmark problem, real cost ($25 vs $30 unit price vs +54% token efficiency), a strengths/weaknesses map, use-case picks, and a dual-vendor strategy.

OpenAI made GPT-5.6 generally available on July 9, 2026, replacing the old "standard + Pro" structure with a three-model lineup: Luna (fast, low-cost, $0.20/$1.20), Terra (balanced, $2/$12, GPT-5.5-class at 40% of Sol's unit price), and Sol (flagship, $5/$30) — prices as revised on July 30, 2026. Sol takes first place on Agents' Last Exam (53.6) and the Coding Agent Index (80), and its 88.8% on TerminalBench 2.1 edges out Claude Fable 5 (86.0%) — yet on production-grade SWE-Bench Pro, Claude Fable 5 leads decisively at 80.0% against Sol's 64.6%. This article covers the differences between the three models, pricing, benchmarks, new features (Programmatic Tool Calling, ChatGPT Work, the full-duplex GPT-Live voice model), availability by ChatGPT plan, a comparison with Claude (Fable 5 / Opus 4.8), and how to choose by use case — all grounded in OpenAI's official announcement and independent benchmarks.

You built on OpenAI, then wanted to try Claude and compare Gemini — and lost hours to the different SDKs, formats, and error handling per provider. An LLM gateway (AI gateway / LLM proxy) is a relay you slot between your app and the providers: it exposes one OpenAI-compatible API to reach every model and takes over the cross-cutting chores — fallback, cost tracking, virtual keys, caching, rate limiting, and observability. This guide covers why you need one, what a gateway really is, the three types (self-hosted proxy = LiteLLM / hosted = OpenRouter / SDK = Vercel AI SDK), how to choose among LiteLLM, OpenRouter, and the Vercel AI SDK, minimal setup code that only swaps the endpoint, and the limits — a hop of latency, the gateway as a new failure point, fees (OpenRouter charges 5.5% on purchases), feature loss, and privacy.

After you build an AI agent, you always hit the same wall: "OK, but is it actually working?" The mechanism for deciding whether a prompt or model change made things better or worse with data instead of gut feel is evals. LLMs produce different output every time for the same input, so exact-match unit tests don't fit. This article covers what evals are, five ways to measure quality (① ground-truth matching ② rule-based checks ③ LLM-as-judge ④ regression testing ⑤ production monitoring), agent-specific evaluation (task success rate, correct tool calls, trajectory, cost), how to start small from 20 failure examples, common pitfalls, and key tools (Anthropic Console/Evals, OpenAI Evals, LangSmith, Langfuse, Ragas) — written for practitioners.

The perennial automation question: "AI agents or RPA?" The answer isn't either/or — choose by role, and the 2026 winning pattern is a hybrid of both. RPA is deterministic "hands" that run a fixed procedure fast and precisely (but break when the screen/spec changes); an AI agent is a probabilistic "brain" that reads the situation and decides (strong on ambiguity and exceptions, but not identical every time). This article covers the operating-principle difference, a comparison table (the reproducibility-vs-resilience trade-off), how to choose (the axis is "can it be fully written as rules?" — yes → RPA, judgment you can't write → AI agent), the 2026 trend (RPA leaders UiPath, Automation Anywhere, Blue Prism going agentic — convergence; the question is no longer "which one" but "where should the reasoning live" = orchestration-first), and the practical answer: a hybrid where the brain (AI agent) handles judgment/orchestration and the hands (RPA) run deterministic execution — don't put an agent where determinism is required, and pair delegated judgment with guardrails and human approval. Based on vendors' official information, with an FAQ.

Can you hand AWS operations to AI? In 2026 you can delegate a lot. AWS itself ships Amazon Q Developer and the Agent Toolkit for AWS (May 2026 — 40+ agent skills + a managed AWS MCP Server + plugins), so AI can reach from IaC generation to resource operations. This guide frames "delegating" in three levels (① code/IaC generation, ② read-oriented ops/investigation, ③ an autonomous agent that actually operates AWS), covers the main tools (Amazon Q Developer, Agent Toolkit, AWS MCP Server, Terraform MCP, Bedrock AgentCore) — including the bring-your-own route of giving Claude Code or Codex the AWS CLI to run "aws" from the shell — the upside (fast IaC, automated triage, cost-optimization ideas, democratized knowledge), and then the real point, the downsides (IAM permission sprawl, over-privilege as a blast-radius amplifier for mistakes/prompt injection, permissions that outlive the task, cost runaway — with real prod-DB-deletion incidents in 2025-26), based on AWS official and security-vendor sources. The key twist: the question isn't "can it?" but "how do you delegate without a runaway or bill explosion" — and AWS itself building IAM guardrails, CloudTrail audit, and sandboxing into the Agent Toolkit shows the shape of the answer. Includes the five principles (least-privilege IAM, human approval for destructive ops, observability, JIT short-lived credentials, sandboxing) and an FAQ.