Back to Explore
NewsVibeFix editorialUpdated Nov 20, 2026

ReSharper 2026.3 Brings Claude Agent and Codex into Visual Studio, with Day-One .NET 11 Support

JetBrains released ReSharper 2026.3 on November 10: Claude Agent and Codex join Junie in Visual Studio's AI Assistant window, all three can run real ReSharper refactorings, and .NET 11 with C# 15 is supported on launch day. The catch is the bill — both newcomers need a JetBrains AI subscription, and bring-your-own-key is not supported.

JetBrains' official ReSharper 2026.3 artwork: the JetBrains logo and “ReSharper 2026.3” beside an operator in goggles with a robotic arm laser-cutting a red gem.

Visual Studio can finally call Claude and Codex directly

On November 10, JetBrains released ReSharper 2026.3. The headline is one sentence long: the AI Assistant tool window inside Visual Studio can now pick Claude Agent and Codex, alongside JetBrains' own Junie. The previous release, ReSharper 2026.2, had just brought Junie into Visual Studio as the extension's first coding agent; one version later, the list of agents in that window has grown from one to three. For .NET developers who live in Visual Studio every day, this means no more switching to a terminal or a separate app to call an agent — writing code, editing code, running terminal commands, and asking questions can all happen inside the editor.

JetBrains' official ReSharper 2026.3 artwork: the JetBrains logo and “ReSharper 2026.3” beside an operator in goggles with a robotic arm laser-cutting a red gem.

It helps to put this in context. Over the past year, the main battlefield for coding agents has been the terminal and standalone apps: Claude Agent lives on the command line, Codex has its own CLI, and editor vendors mostly connected to them rather than building their own. JetBrains' choice says a lot: instead of betting that its own agent could beat everyone, it turned ReSharper into a host that can hold three agents at once. Junie is still the home-grown one, but Claude Agent and Codex are now options on equal footing. That is a concession, and a pragmatic one — developers already have agents they like, and an editor should let them in rather than force a switch.

What actually changed

The first change is the entry point. Install or update to ReSharper 2026.3, make sure AI Assistant is selected in the installer, and the agent list in the AI Assistant tool window will show Claude Agent, Codex, and Junie. Pick one, send a first prompt, and you are off. The announcement puts it plainly: you can ask your chosen agent to write and edit code, run terminal commands, or explain unfamiliar parts of your solution — all without leaving Visual Studio. The first time you use Claude Agent or Codex, ReSharper asks for your consent before installing the selected agent, since it is developed by a third party. If Node.js is not installed on your machine, ReSharper installs that too.

The second change is how they connect. All three agents connect through the Agent Client Protocol (ACP), an open protocol for connecting coding agents to editors. The protocol itself is not news; the news is that JetBrains is willing to use an open protocol to bring in other vendors' agents instead of keeping only a home-grown channel. That is good for developers: the more open the protocol, the cheaper it is to switch agents later, and the lower the risk of being locked to a single model vendor by your editor.

The third change deserves a closer look, and it is the easiest to miss in this release: all three agents can use ReSharper's refactoring skill. That means when an agent needs to rename a symbol or move a type to another namespace, it runs a real refactoring instead of editing files one by one. The difference is concrete: text replacement often misses references and breaks the build, while a refactoring follows the reference chain and changes everything in one pass. For vibe coding, this is exactly where agents most often stumble — they can write code that looks right but leave half-updated references behind when renaming or moving things. Letting the agent use the editor's own refactoring for the most error-prone step hands the hardest job to the party that understands the code structure best. That design choice is more worth remembering than "two more model names".

Let us get the money question out of the way first

Three agents being available does not mean three agents are free to try. Claude Agent and Codex require a JetBrains AI subscription and consume your JetBrains AI quota; Junie does too, unless you use the free Junie Lite tier. Junie Lite is the free tier launched alongside this release: anyone signed in with a JetBrains Account can use it, even without a ReSharper license. It is the on-ramp for people who have not bought ReSharper yet — try Junie for free first, then decide whether to pay.

There are two limits to know before you install, so you do not discover them after the fact. First, your own Claude or ChatGPT subscription does not work here: the announcement states clearly that this release does not support bring-your-own-key (BYOK), and ReSharper does not use personal Claude or ChatGPT subscriptions. If you want Claude Agent, you go through the JetBrains AI subscription and quota. Second, JetBrains AI Pro trials are gone: ReSharper no longer offers them, though an already-started trial can run until it ends. New users get no trial. In other words, JetBrains has done the math out in the open: the only free tier is Junie Lite, and everything else runs on a subscription.

One more line sits at the end of the announcement in small print: regional restrictions on the availability of AI features in Mainland China apply. It is not elaborated, but for developers in China it is something to check first: what the feature list says is not necessarily all available where your account is. Check what is actually available in your region before you install, rather than finding a greyed-out button when you need it.

Day-one support for .NET 11 and C# 15

The agents are the headline, but for .NET developers something else happened the same day that may matter more in practice: Microsoft launched .NET 11 at .NET Conf 2026, and ReSharper 2026.3 already supports it. JetBrains makes a point of writing "Microsoft launched .NET 11 today" — you do not wait for the next version to get support; you get the new syntax on upgrade day.

C# 15 support covers union types, closed class hierarchies, collection expression arguments, extension indexers, and labeled break and continue statements. Code analysis, completion, navigation, and other features work with the new syntax, and new inspections, quick-fixes, and context actions help you adopt it in existing code. The memory safety redesign preview is also recognized: ReSharper understands the new safe modifier and unsafe expressions, highlights missing or redundant unsafe contexts, and offers quick-fixes for them. On the decompiler side, .NET 11's runtime-async methods now show as ordinary async methods, and the IL Viewer marks them with the async keyword — runtime-async itself is still a preview feature in .NET 11.

The test runner has kept up with new framework versions too: it supports xUnit v3 4.0 and NUnit 5.0, test discovery is more reliable for MSTest 4 projects, and it can read Playwright settings such as LaunchOptions from your .runsettings file, so you can run browser tests with the browser window visible. Microsoft.Testing.Platform support is now enabled by default, so ReSharper finds tests for frameworks like TUnit without any setting changes. On the C++ side there is new material as well: a Class Layout view shows how a class is arranged in memory, format string checks cover std::format, std::print, and the {fmt} library, and there are new inspections for Unreal Engine's structured logging macros and CUDA.

How to get started after installing

The announcement's setup path has three steps, and it is worth following them as written. First, install or update to ReSharper 2026.3, with AI Assistant selected in the installer. If you finish installing and cannot find the AI Assistant tool window, do not reinstall everything — run the installer again and choose "Install or Repair" next to AI Assistant. Second, open the AI Assistant tool window, sign in to your JetBrains Account if asked, and choose Claude Agent or Codex from the agent list. Third, send your first prompt.

The first use of Claude Agent or Codex adds one confirmation step: ReSharper asks for your consent before installing the selected agent, on the grounds that it is developed by a third party. If Node.js is missing from the machine, ReSharper installs it as well, so you do not have to fetch it yourself. The flow is deliberately restrained: nothing is installed by default, nothing third-party is installed silently, and every step asks you to say yes. That matters in corporate environments, where machines often may not install third-party components freely — asking first gives compliance one less reason to object.

One warning about the free tier: Junie Lite needs no ReSharper license, but "free" does not mean "unlimited". The announcement only says it is a free tier available to anyone signed in with a JetBrains Account; it gives no quota cap or rate limit. Before making it your main tool, try it on one or two real tasks and see whether the free tier's response speed and output quality are enough for you, then decide whether to upgrade to a paid subscription. The right way to use a free tier is to find its ceiling, not to assume it has none.

Junie, Claude Agent, or Codex — how to choose

There are now three options in the window. Which one? The announcement gives no recommendation, but the known facts suggest a decision path. Start with cost: Junie Lite is free, while full Junie, Claude Agent, and Codex all consume JetBrains AI quota. If you have not paid anything yet, Junie Lite is the zero-cost starting point; if you already pay for a JetBrains AI subscription, the marginal cost of all three is the same, and the choice comes down to results.

Then look at capability boundaries: all three agents can write code, edit code, run terminal commands, explain code, and call ReSharper's refactoring skill. The announcement says nothing about differences between them on any of these, and gives no hint about which is better at what. The official position is effectively "all three can do the job" — the difference is not in the feature list but in how each feels when you actually use it. Claude Agent runs on Anthropic's models, Codex on OpenAI's, Junie on JetBrains' own models and orchestration. Differences in model character — who is more cautious, who rewrites more aggressively — are not mentioned at all in the announcement. You can only find out by trying them on the same task.

A practical suggestion: do not bet everything on one of them up front. Take a real task you need to do anyway — adding a small feature to an old module, or renaming a type that is referenced everywhere — and let each of the three agents do it once. Compare three things: whether the result compiles on the first try, whether the references were updated cleanly, and whether the change stayed within the scope you asked for. Rename tasks are the most revealing, because they test directly whether the agent used the refactoring skill or edited files one by one. After three trials, you will know. That beats reading any benchmark, because benchmarks use someone else's code and you are testing your own.

Beyond the agents, a batch of everyday improvements

Setting the agents aside, ReSharper 2026.3 carries a set of improvements you will meet while writing code every day, and they deserve a mention. On code analysis, ReSharper now suggests collection expressions in more places, including lists and immutable builders filled with foreach loops, AddRange, or individual Add calls. New inspections catch a few easy mistakes: character values used by mistake as array lengths or indices, culture-sensitive operations, implicit string sorting, and using statements that dispose of a task instead of its result. Another inspection suggests tuple syntax instead of verbose ValueTuple creation. Parameter information now groups overloads that share the same leading parameters, so long overload lists are easier to scan while you write a call.

None of these is dramatic on its own, but together they say something: JetBrains has not put all its effort into agents — traditional static analysis and coding assistance are still being polished. For someone who writes code all day, an agent might get called a few times a day, while inspections and completion run every minute. Making the latter a little better is felt more directly than adding one more model to the former. It is also a way to judge a release: headlines alone tend to overstate things, and the everyday improvements beyond the headline show how solid the version really is.

One more easily missed point: this release is ReSharper, ReSharper C++, and dotPeek together as 2026.3. The decompiler's runtime-async display improvements are in dotPeek 2026.3 as well, and the C++ Class Layout view and format string checks are in ReSharper C++ 2026.3. If you write across languages, 2026.3 is a bundled upgrade, not an update for C# developers only.

What this means if you write .NET every day

First, the good news. If Visual Studio and .NET are your home base, this update handles two things at once: agents without leaving the editor, and .NET 11 on launch day. The refactoring skill matters most for people who often ask agents to change old code — renaming and moving types used to produce compile errors, and now at least those two jobs go through the editor's own path, which is far more reliable. The Junie Lite free tier also lowers the barrier: you can step inside and try before buying.

Now the cooler-headed part. First, of the three agents, two need a subscription and one has a limited free tier, so the JetBrains AI quota becomes the new ceiling. If you used to run Claude Agent in the terminal on your own Claude subscription, the cost and quota were a different account; moving into ReSharper means redoing that math, and this version does not support bring-your-own-key. Second, the agents can be called in, but how well they work once inside is unmeasured in the announcement — no benchmark scores, no comparisons, no task success rates. The differences between the three agents on the same task are yours to discover. Third, AI Assistant and JetBrains' new Air layer run side by side: Air is the agent-running layer JetBrains is shipping in 2026.3, while AI Assistant remains available as a separate plugin until early 2027. Two lines coexisting means more choice in the short term, but they will have to converge eventually, and it is not yet visible which line wins.

One judgment is worth writing down: JetBrains' real bet here is not Junie, it is the host position. Models change generations, agents come and go, and if the editor becomes the place every agent can enter, it never has to bet on which model wins. That logic is friendly to developers — you like Claude today, Codex tomorrow, and the editor stays the same. But it has a price: the more open the host, the more important the control over subscriptions and quotas becomes, and the JetBrains AI subscription is turning into the new lock. Trying Junie Lite for free is fine, but making Claude Agent and Codex your main tools means first accepting that subscription math.

Finally, who should upgrade now? If Visual Studio and .NET 11 are your daily work, the case is strong: three agents to choose from plus day-one .NET 11 support, two benefits in one upgrade. If you are still on .NET 8 or .NET 10 and in no rush for .NET 11, the attraction narrows to the agents alone — try Junie Lite for free first, then decide whether to pay for Claude Agent and Codex. And if your daily flow already runs Claude Agent or Codex in the terminal, with a subscription and workflow that work, there is no hurry: the ReSharper versions do not support bring-your-own-key yet, so moving in means changing how you pay. Wait and see whether BYOK arrives in a later version before deciding.

What is still unanswered

The announcement leaves several practical questions open. First, how well do the three agents actually perform inside Visual Studio? It says what they can do, not how well they do it, and gives no task success rates or comparisons. Second, how exactly is the quota for Claude Agent and Codex calculated? It only says they consume JetBrains AI quota, with no numbers — the cost of different tasks is something you find out by using them. Third, what is the long-term relationship between Air and AI Assistant? Air is just rolling out in 2026.3, AI Assistant is kept until early 2027, and whether they merge or one is picked over the other has not been said. Fourth, ReSharper C++ 2026.3 shipped the same day, with the Class Layout view and format string checks as its headlines — C++ developers should read that separate announcement rather than looking for the details here.

One sentence to sum up: the most important thing in ReSharper 2026.3 is not two more names on a list, it is the admission that an editor's future is to be a good host — letting the agents developers like come in and do the work, and letting them use the editor's own refactoring for the jobs most likely to go wrong. Subscriptions and quotas are the part of that deal that needs to be worked out in advance, before you install it and send your first prompt, only to find the quota runs out first.

Views 16Comments 0

Comments (0)

ME
0/1000
Loading comments...

Sources

Browse projectsPublish your project

Related articles

A developer with glasses types on a laptop at a wooden desk, an external monitor to the left showing C code on a dark background, a small plant on the desk
News
Microsoft's Coding Model Stops Playing Second Fiddle: MAI Code 2 Scores 96% of Opus 5.5 for 5% of the Cost — Read the Three Caveats First

On 17 November 2026, Microsoft launched MAI Code 2 and Code 2 Flash: the larger model has 77 billion active parameters and a one-million-token context, averaging 63.2% across four benchmarks — close to GPT-6.1 Sol and Claude Opus 5.5 — at $0.309 per task. Read the three caveats first: promotional pricing, the lowest score of the three, and no long-horizon work.

Developer WorkflowProduct LaunchModel Updates
A developer types at a dual-monitor workstation; the left screen shows a dark code editor, the right a testing dashboard with green checkmarks and bar charts
News
Did Your Agent Get Better or Worse? AWS Open-Sources UAEF to Measure Agents from Every Framework with One Ruler

On 11 November 2026, AWS open-sourced UAEF: adapters turn traces from LangGraph, LangChain, Strands and Bedrock AgentCore into one format, then 36 metrics across 7 dimensions score them, flagging regressions past 5%. The judge model is hardened too — scores must cite verbatim evidence, and never decide safety.

Developer WorkflowProduct LaunchTesting & Quality
Against a black background, a conductor seen from behind raises a baton, with swirling trails of light circling beneath him like an orchestra being conducted
News
One API for 45 Models, and It Switches Providers When One Fails: Perplexity's Orchestrator Wants to Be the Switchboard

On 17 November 2026, Perplexity launched the Orchestrator API: one interface reaching OpenAI, Anthropic, Google, xAI and open-weight models such as GLM, Kimi and DeepSeek, billed at the answering model's published rate with no platform fee. When one provider fails, the same model retries on another deployment, and preview routing assigns requests to models by task.

Developer WorkflowProduct LaunchModel Updates