What Shipped Since You Installed Claude Code: Opus 4.8, Dynamic Workflows, The Agent View, And The Q2 2026 Surface

A companion to the Claude Code workflow series. If you set up Claude Code earlier in 2026 and have not looked at the changelog since, this is the catch-up map. Everything here is from the official docs as of June 2026.

You installed Claude Code, learned the slash commands, built a CLAUDE.md, and got into a rhythm. Then a couple of months passed. The tool did not sit still. Spring 2026 brought a new default model, a way to run hundreds of agents in the background, a dashboard for sessions you are not watching, a goal-tracking mode, a review command that finds real bugs, and the ability to drive the whole thing from your phone.

This post is the organized version of "what did I miss." It is grouped by what you would actually reach for, with the version each thing landed in so you can check your own install with claude --version.

The model: Opus 4.8 and fast mode

The headline release was Opus 4.8 (v2.1.154, late May 2026). On the Anthropic API and on Max, Team Premium, and Enterprise pay-as-you-go tiers, opus now resolves to Opus 4.8 and it is the default. If you are on Pro or Team Standard, your default is still Sonnet 4.6, and you switch with /model. That tier distinction trips people up, so it is worth knowing which model you are actually getting.

Two things came with it that change day-to-day use:

• Fast mode. The same Opus 4.8 model runs at up to 2.5x the output speed for 2x the standard token rate (fast pricing is $10 and $50 per million input and output tokens, against $5 and $25 standard). Max-plan Claude Code users default to fast mode on Opus 4.8. You feel it most on long, chatty sessions.
• A longer effort ladder. The effort levels are now low, medium, high, xhigh, and max. Opus 4.8 defaults to high, and /effort xhigh is there for the genuinely hard problems where you want deeper reasoning and do not mind the token spend. Hooks and Bash commands can read the active level through $CLAUDE_EFFORT.

A quieter change shipped alongside: a leaner system prompt is now the default for current models, which leaves a bit more of the context window for your actual work.

Dynamic workflows: orchestrate agents from one prompt

This is the biggest new capability. Dynamic workflows (/workflows, also from v2.1.154, research preview) let Claude write a small JavaScript orchestration script that fans work out across tens to hundreds of background subagents, up to sixteen running at once, while your main session stays responsive. You describe the shape of the work (decompose, run in parallel, verify, synthesize) and the script runs it deterministically.

Two related pieces:

• The in-prompt trigger keyword was renamed from "workflow" to "ultracode" (v2.1.160). Asking in plain language ("run a workflow to audit this") still works; the literal keyword is what changed. Separately, /effort ultracode is an effort-menu setting that sends xhigh to the model and auto-orchestrates a workflow for each substantial task in the session.
• /deep-research is a bundled workflow: it fans out web searches across angles, fetches and cross-checks sources, votes on each claim, and returns a cited report with the unsupported claims filtered out. It needs the WebSearch tool.

If you have been doing big migrations or audits by hand in one long session, this is the feature to learn next.

The agent view and background sessions

Once you can spawn a lot of work, you need somewhere to watch it. The agent view (claude agents, v2.1.139) is a single dashboard for every background session: dispatch new ones, see what is running, what needs your input, and what is done, and attach to a session only when it actually needs you. There is a --json mode for scripting and --cwd to scope the list to a directory.

Backing that view, background sessions became first-class:

• claude --bg "<prompt>" starts a session that goes straight to the background (it accepts --name, --agent, --model, and --permission-mode).
• claude --bg --exec '<command>' runs a shell command as a PTY-backed background job with no model involved, the same as typing ! <command> in the agent view.
• /bg inside a running session pushes it into the background, and these sessions show up in /resume marked bg.

The mental model: your terminal becomes a launcher and a monitor, not a place you sit and wait.

/goal: keep working until the condition is met

/goal (v2.1.139) sets a completion condition and Claude keeps working across turns until a small fast model confirms the goal is actually met, then it clears on its own. You get a live overlay of elapsed time, turns, and tokens. It works in interactive sessions, in -p, and over Remote Control. Use it for "do not stop until the test suite is green" style tasks where you would otherwise keep nudging it turn after turn.

/code-review and the /simplify split

This one changed twice, so here is the current truth. The command that used to be /simplify was renamed to /code-review (v2.1.147). /code-review reviews the current diff for correctness bugs plus reuse, simplification, and efficiency cleanups at a chosen effort level. --comment posts findings as inline GitHub PR comments, --fix applies them to your working tree, and /code-review ultra kicks off the deeper multi-agent cloud review.

The catch: /simplify did not disappear, and it is not just an alias. As of v2.1.154 it runs a separate, cleanup-only review (reuse, simplification, efficiency, altitude) that applies fixes but deliberately does not hunt for bugs. So the rule is simple: /code-review --fix when you want bugs found and fixed, /simplify when you only want the code tidied. There was a brief window where /simplify just called /code-review --fix; that was reverted, so do not rely on it.

Remote control: the session leaves the terminal

You can now drive a running local session from your phone, a browser, or VS Code with Remote Control (v2.1.51+). The session stays on your machine with all your files and MCP servers; the phone is just a window. Start with claude --remote-control (or /remote-control inside a session), then connect by URL, QR code, or the session list at claude.ai/code. Mobile push notifications (v2.1.110+) tell you when a long task finishes or needs a decision. It makes outbound HTTPS calls only and never opens an inbound port, so there is no tunnel into your laptop.

Channels push events the other direction, from iMessage, Telegram, Discord, or a CI webhook into your running session, and Dispatch lets you start a session from a text on your phone. The full setup for all three is in the dedicated post: Remote Control + Channels.

Plugins, skills, and a calmer Auto Mode

A few smaller but useful changes:

• Plugins load automatically from .claude/skills with no marketplace required (v2.1.157), /reload-skills (v2.1.152) picks up newly installed skills without restarting the session, and claude plugin init <name> scaffolds a new plugin for you.
• Auto Mode no longer requires opt-in consent (v2.1.152), and it became available on Bedrock, Vertex, and Foundry for Opus 4.7 and 4.8. The boundaries still matter (production deploys and high-blast-radius operations should stay gated), but the routine middle is now the default rather than a flag you have to remember.

Codex kept pace, and the two pair well

If you also run OpenAI's Codex, it moved just as fast, and the interesting part is how well the two fit together now.

On the model side, Codex defaults to gpt-5.5 for ChatGPT sign-in sessions, with gpt-5.4-mini as the cheap option for exploration and subagents. API-key sessions may still point at gpt-5.2-codex, so check rather than assume. Codex also picked up many of the same shapes Claude Code has: a Goal Mode, subagents, permission profiles in place of a blanket full-auto, an in-app browser and computer use, hooks, plugins, MCP, and its own remote control.

The reason to run both is that their strengths are complementary, not redundant:

• Plan in Claude Code, execute in Codex. Claude reasons well about ambiguous, big-picture design; Codex is strong at deterministic, mechanical execution across many files with each step checked against tests. Write the plan in Claude, hand it to Codex.
• Use the official Codex plugin for Claude Code. OpenAI shipped a plugin (codex-plugin-cc) that brings Codex into a Claude Code session for a standard review, a more skeptical adversarial review, or a handoff to a different model for a second pass. A model reviewing its own work misses things a different model catches.
• Codex executes unattended, Claude reviews. Codex runs the batch job and opens a PR; Claude Code reviews the diff for consequences. Throughput from one, judgment from the other.
• Route by task shape. Claude for ambiguous, visual, or multi-agent work; Codex for tight, well-scoped, cost-conscious tasks. The two-CLI workflow post goes deeper on the split.

Pricing and exact model defaults drift on both sides, so confirm current numbers before you wire any of this into a routine.

Catch up in twenty minutes

If you only have a short sitting, do these in order:

1. claude --version, then update if you are behind. A lot below needs v2.1.154 or later.
2. Run /powerup once. It surfaces features you have not turned on yet.
3. Make your next review a /code-review instead of an eyeball pass, and try --fix.
4. Turn on Remote Control with claude --remote-control, connect your phone, and enable "Push when Claude decides" in /config.
5. Point /deep-research at a real question to see a dynamic workflow run end to end.
6. Glance at claude agents so the dashboard is familiar before you actually need it.

That is the difference between "I installed it months ago" and "I am current."

If you are using these tools to run lean client work or a one-person shop, the same "small stack, big leverage" idea runs through my book The $20 Dollar Agency (one of the $9.99 Kindle titles), which is about delivering real work without a real budget. The Q2 Claude Code surface is most of that stack in one CLI.

What to watch next

Remote Control, Channels, and dynamic workflows are all research-preview features, which means they are still changing release over release. The honest way to stay current is to skim the official changelog every few weeks. It is terse, dated, and the single source of truth. Any roundup, including this one, is a snapshot.

Related reading

• CLI installed, now what? — the workflow foundations these features build on.
• Remote Control + Channels — the full setup for running a session from your phone.
• Two-CLI workflow — Codex and Claude Code side by side.
• /loop vs /schedule — in-session polling vs scheduled runs, which the agent view now complements.
• AI Terminal Kickstart — the install primer, if you are setting up a fresh machine.

Fact-check notes and sources

All version numbers and behaviors are from official Anthropic and Claude Code sources as of June 2026. Research-preview features can change between releases.

• Official Claude Code docs — Changelog, Fast mode, Dynamic workflows, Agent view, /goal, /code-review, Remote Control, Channels, and model configuration.
• Model announcement — Claude Opus 4.8.
• Codex side — Codex models, Codex changelog, Codex CLI releases, and the Codex plugin for Claude Code announcement.

This post is informational, not engineering-consulting advice. Commands, flags, version numbers, and prices reflect the Claude Code and Codex surfaces as of June 2026 and will drift; re-verify against the official changelogs before depending on any specific detail. Mentions of Anthropic, Claude Code, OpenAI, and Codex are nominative fair use. No affiliation is implied.