an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions
HTML agent-harnessagent-orchestrationagentsai-agentsautonomous-agentsclaude-codecodexcoding-agentsdesktop-appdeveloper-toolselectrongemini-cliharnessharness-engineeringlocal-firstmemorymulti-agentopencodeorchestrationtypescript
<!doctype html>
<html lang="en">
<head>
<!-- Google tag (gtag.js), Google Ads AW-18388216863. Loaded only when Do Not Track is
off, so it follows the same rule as the PostHog block on this site. -->
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){ dataLayer.push(arguments); }
if (navigator.doNotTrack !== '1') {
var gtagScript = document.createElement('script');
gtagScript.async = true;
gtagScript.src = 'https://www.googletagmanager.com/gtag/js?id=AW-18388216863';
document.head.appendChild(gtagScript);
gtag('js', new Date());
gtag('config', 'AW-18388216863');
}
</script>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Evaluating AI Agent Reliability: How to Measure What You Can Trust — Munder Difflin Blog</title>
<meta name="description" content="How to measure whether an AI agent is reliable enough to trust: reliability thresholds, why benchmarks overstate, and evaluating on your own codebase." />
<link rel="canonical" href="https://munderdiffl.in/blog/evaluating-ai-agent-reliability/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
<link rel="apple-touch-icon" href="https://munderdiffl.in/logo.png" />
<meta name="theme-color" content="#F5F2E8" />
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="Evaluating AI Agent Reliability: How to Measure What You Can Trust — Munder Difflin Blog" />
<meta property="og:description" content="How to measure whether an AI agent is reliable enough to trust: reliability thresholds, why benchmarks overstate, and evaluating on your own codebase." />
<meta property="og:url" content="https://munderdiffl.in/blog/evaluating-ai-agent-reliability/" />
<meta property="og:image" content="https://munderdiffl.in/blog/assets/media/evaluating-ai-agent-reliability/hero.png" />
<meta property="og:image:width" content="1600" />
<meta property="og:image:height" content="900" />
<meta property="og:image:alt" content="Illustration: Evaluating AI Agent Reliability: How to Measure What You Can Trust" />
<meta property="og:locale" content="en_US" />
<meta property="article:published_time" content="2026-06-04T00:00:00.000Z" />
<meta property="article:author" content="Chaitanya Giri" />
<meta property="article:tag" content="Evaluation" />
<meta property="article:tag" content="Autonomous" />
<meta property="article:tag" content="Agentic AI" />
<meta property="article:tag" content="Guardrails" />
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="Evaluating AI Agent Reliability: How to Measure What You Can Trust — Munder Difflin Blog" />
<meta name="twitter:description" content="How to measure whether an AI agent is reliable enough to trust: reliability thresholds, why benchmarks overstate, and evaluating on your own codebase." />
<meta name="twitter:image" content="https://munderdiffl.in/blog/assets/media/evaluating-ai-agent-reliability/hero.png" />
<meta name="twitter:image:alt" content="Illustration: Evaluating AI Agent Reliability: How to Measure What You Can Trust" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
<link rel="preconnect" href="https://fonts.googleapis.com" />
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
<link href="https://fonts.googleapis.com/css2?family=Fraunces:opsz,[email protected],500;9..144,600;9..144,700;9..144,900&family=Source+Serif+4:opsz,[email protected],400;8..60,600;8..60,700&family=Space+Grotesk:wght@500;600;700&family=JetBrains+Mono:wght@400;600&display=swap" rel="stylesheet" />
<link rel="stylesheet" href="/blog/assets/blog-press.css" />
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "Evaluating AI Agent Reliability: How to Measure What You Can Trust",
"description": "How to measure whether an AI agent is reliable enough to trust: reliability thresholds, why benchmarks overstate, and evaluating on your own codebase.",
"image": ["https://munderdiffl.in/blog/assets/media/evaluating-ai-agent-reliability/hero.png"],
"datePublished": "2026-06-04T00:00:00.000Z",
"dateModified": "2026-06-04T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
"name": "Munder Difflin",
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/evaluating-ai-agent-reliability/" },
"keywords": "evaluating ai agent reliability, measure ai agent reliability, ai agent evaluation, agent eval harness, ai agent benchmarks limits",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/evaluating-ai-agent-reliability/"
}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BreadcrumbList",
"itemListElement": [
{ "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
{ "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
{ "@type": "ListItem", "position": 3, "name": "Evaluating AI Agent Reliability: How to Measure What You Can Trust", "item": "https://munderdiffl.in/blog/evaluating-ai-agent-reliability/" }
]
}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{ "@type": "Question", "name": "How do you measure if an AI agent is reliable enough to trust?", "acceptedAnswer": { "@type": "Answer", "text": "Pick a reliability bar and measure against it on tasks like yours. A 50%-reliable agent is a demo; something you'd run unattended needs to be right far more often. METR tracks both 50%- and 80%-reliability horizons for exactly this reason. Then evaluate on your own codebase, not a public benchmark, because curated benchmarks are an optimistic ceiling." } },
{ "@type": "Question", "name": "Are benchmarks like SWE-bench enough to judge a coding agent?", "acceptedAnswer": { "@type": "Answer", "text": "No. SWE-bench Verified is a human-curated set of public GitHub issues confirmed to be solvable with fair tests — useful for comparing models, but an optimistic ceiling. Your private, underspecified, weakly-tested codebase is the harder case the benchmark filters out, so real-world reliability is consistently lower than the headline score." } },
{ "@type": "Question", "name": "What should an AI agent eval harness measure?", "acceptedAnswer": { "@type": "Answer", "text": "Both correctness and the things that actually break agents over long runs. Use verifiable checks (your test suite) as ground truth for 'did it work,' an LLM-as-judge for fuzzier qualities like readability or security (with care — judges are themselves unreliable), and track tool-use consistency, error recovery, and context management. Run it in CI on every change to agent logic, and calibrate against human review on a small sample." } }
]
}
</script>
<script>
// Anonymous blog analytics (PostHog) — same posture as the main site and the
// app's TELEMETRY.md: public write-only token ('' = inert), anonymous-only
// (person_profiles 'never'), no autocapture/recording, DNT respected.
(function () {
var PH_KEY = 'phc_qJzsokwcbtTGHemnvywR3nSqivNNgMjHeNVEWkZNydFB';
if (!PH_KEY || navigator.doNotTrack === '1') return;
var s = document.createElement('script');
s.async = true;
s.src = 'https://us-assets.i.posthog.com/static/array.js';
s.onload = function () {
if (!window.posthog) return;
window.posthog.init(PH_KEY, {
api_host: 'https://us.i.posthog.com',
person_profiles: 'never',
autocapture: false,
before_send: function (event) {
// PostHog fills $ip from the connection unless the event carries one.
// Null it here and disable the GeoIP lookup so no IP, city, postal
// code or lat/long is ever derived or stored for this page view.
if (event && event.properties) {
event.properties.$ip = null;
event.properties.$geoip_disable = true;
}
return event;
},
disable_session_recording: true,
respect_dnt: true
});
};
document.head.appendChild(s);
})();
</script>
</head>
<body class="theme-press">
<a class="skip-link" href="#main">Skip to content</a>
<nav id="nav">
<a class="brand" href="https://munderdiffl.in" aria-label="Munder Difflin home">
<img class="brand-logo" src="https://munderdiffl.in/logo.png" width="1318" height="840" alt="Munder Difflin, Inc. — Multi-Agent Harness" />
</a>
<span class="spacer"></span>
<div class="links">
<a href="/blog/">Latest</a>
<a href="/blog/topics/">Topics</a>
<a href="/blog/about/">About</a>
<a href="https://munderdiffl.in">↗ Product</a>
</div>
<div class="nav-actions">
<a class="btn sm" href="https://munderdiffl.in/blog/feed.xml" aria-label="RSS feed">⤵ RSS</a>
<a class="btn sm primary" href="https://munderdiffl.in/#install">⤓ Download</a>
</div>
</nav>
<main id="main">
<article class="t-guides">
<header class="post-head wrap">
<nav class="crumbs" aria-label="Breadcrumb">
<a href="/blog/">Blog</a>
<span>/</span>
<a href="/blog/topics/guides/">Guides</a>
</nav>
<h1>Evaluating AI Agent Reliability: How to Measure What You Can Trust</h1>
<p class="dek">How to measure whether an AI agent is reliable enough to trust: reliability thresholds, why benchmarks overstate, and evaluating on your own codebase.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-06-04">Jun 4, 2026</time>
<span class="dot" aria-hidden="true">·</span>
<span>5 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
<div class="hero-wrap wrap"><figure class="fig fig-hero"><img src="/blog/assets/media/evaluating-ai-agent-reliability/hero.png" alt="Illustration: Evaluating AI Agent Reliability: How to Measure What You Can Trust" loading="lazy" decoding="async" /></figure></div>
<div class="post-wrap wrap">
<div class="prose">
<div class="callout tldr"><span class="ic">TL;DR</span><p>"Reliable enough" is a
<strong>measurement, not a vibe</strong>. Pick a reliability bar — a 50%-success agent is a demo; something
you trust unattended needs far higher. Treat public benchmarks like SWE-bench Verified as an
<strong>optimistic ceiling</strong>, not a verdict on your codebase. Measure the things that actually
break agents — <strong>tool-use consistency, error recovery, context management</strong> — not just
final-answer accuracy. Then build a lightweight <strong>eval harness on your own work</strong>: verifiable
checks as ground truth, LLM-as-judge for the fuzzy parts (carefully), regression eval in CI, and human
calibration on a sample.</p></div>
<p>Before you hand an agent real work, the question isn’t “is it smart?” — it’s “how often is it right, on
tasks like mine, at a bar I can live with?” That’s an evaluation problem, and most teams skip it, trust a
leaderboard, and get surprised. This is a practical guide to measuring agent reliability honestly.</p>
<blockquote>
<p><strong>On sourcing.</strong> Figures below are linked and current to mid-2026. Benchmark numbers move weekly and
depend heavily on scaffold and setup — treat them as directional, and prefer evals you run yourself.</p>
</blockquote>
<h2 id="reliability-is-a-number-you-have-to-pick" tabindex="-1">Reliability is a number you have to pick <a class="anchor" href="#reliability-is-a-number-you-have-to-pick" aria-hidden="true">#</a></h2>
<p>“Reliable” is meaningless without a threshold. The most useful framing comes from
<a href="https://metr.org/time-horizons/">METR</a>, which measures an agent’s “task-completion time horizon” at a
<em>chosen</em> reliability — and reports it at <strong>both 50% and 80%</strong>. The gap between those two is the whole
point: an agent might handle a long task at coin-flip odds while only reliably finishing a much shorter
one. A 50%-success rate is fine for a demo you supervise; for anything you’d run unattended, you want a
high bar, and you should know which bar you’re quoting. Decide your threshold <em>before</em> you measure, or
you’ll unconsciously pick the number that flatters the agent.</p>
<h2 id="benchmarks-are-a-ceiling-not-a-verdict" tabindex="-1">Benchmarks are a ceiling, not a verdict <a class="anchor" href="#benchmarks-are-a-ceiling-not-a-verdict" aria-hidden="true">#</a></h2>
<p>Public benchmarks are useful for comparing models and useless as a promise about your code. The standard,
<a href="https://www.demandsphere.com/research/demandsphere-radar/ai-frontier-model-tracker/benchmarks/swe-bench/">SWE-bench Verified</a>,
is a set of real GitHub issues <strong>human-validated to be solvable with fair tests</strong> — top models score
around 80% on it. But that curation is exactly what makes it optimistic: it filters out the unsolvable,
the ambiguous, and the badly-tested — i.e., most of a real backlog. Your private, underspecified,
weakly-tested codebase is the harder case the benchmark removed. As benchmark roundups themselves now
<a href="https://kili-technology.com/blog/ai-benchmarks-guide-the-top-evaluations-in-2026-and-why-theyre-not-enough">caution</a>,
a leaderboard score is a starting point, not a verdict — real-world reliability is consistently lower.</p>
<p>The takeaway isn’t “ignore benchmarks.” It’s: use them to <em>rank</em>, then re-measure on <em>your</em> work before
you trust anything.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/evaluating-ai-agent-reliability/note-1.png" alt="Hand-drawn sketch from “Evaluating AI Agent Reliability: How to Measure What You Can Trust”" loading="lazy" decoding="async" /></figure>
<h2 id="measure-the-things-that-actually-break" tabindex="-1">Measure the things that actually break <a class="anchor" href="#measure-the-things-that-actually-break" aria-hidden="true">#</a></h2>
<p>Final-answer accuracy hides the failure modes that matter for an autonomous agent. What predicts
real reliability is the stuff that compounds over a long run:</p>
<ul>
<li><strong>Per-step reliability.</strong> Reliability is multiplicative: a 95%-per-step agent is only ~60% reliable
across a 10-step chain. Measure the chain, not the step.</li>
<li><strong>Tool-use consistency.</strong> Does it call the right tool with the right arguments, repeatedly?</li>
<li><strong>Error recovery.</strong> When a command fails, does it diagnose and adapt, or loop and flail? This is
<a href="/blog/recovering-from-agent-failures/">failure-recovery</a> under measurement.</li>
<li><strong>Context management.</strong> Over a long session, does it stay on the original goal or drift?</li>
</ul>
<p>Production eval frameworks have converged on this multi-dimensional view — see, e.g., this
<a href="https://towardsdatascience.com/building-an-evaluation-harness-for-production-ai-agents-a-12-metric-framework-from-100-deployments/">12-metric harness distilled from 100+ deployments</a>.
The headline accuracy is one column; the ones that catch unattended failures are the others.</p>
<h2 id="build-an-eval-harness-on-your-own-work" tabindex="-1">Build an eval harness on your own work <a class="anchor" href="#build-an-eval-harness-on-your-own-work" aria-hidden="true">#</a></h2>
<p>You don’t need a research lab — you need a small, repeatable harness over tasks that look like yours.
A practical recipe, drawn from current
<a href="https://www.adaline.ai/blog/complete-guide-llm-ai-agent-evaluation-2026">agent-evaluation practice</a>:</p>
<ul>
<li><strong>Verifiable checks as ground truth.</strong> For “did it work,” nothing beats your <strong>test suite</strong>. A task
with a concrete pass/fail check is the cleanest possible eval signal — and the same tests that grade the
agent are the bar you should <a href="/blog/claude-code-automation-while-you-sleep/">make it work toward</a> in the
first place.</li>
<li><strong>LLM-as-judge for the fuzzy parts — carefully.</strong> For readability, security, or “is this a good fix,”
an LLM grader scales where humans can’t, and it pairs naturally with having
<a href="/blog/how-ai-agents-verify-their-own-work/">agents verify their own work</a>. But judges are themselves
unreliable: studies find <a href="https://arxiv.org/pdf/2603.05399">no judge is uniformly trustworthy</a>, with
verdicts swinging on formatting, paraphrasing, and verbosity. Use rubrics, and don’t treat a judge
score as gospel.</li>
<li><strong>Regression eval in CI.</strong> Run the harness on every change to agent logic or prompts and compare to the
last run, the way you’d run a test suite — flag improvements and regressions per metric, not just an
overall pass.</li>
<li><strong>Human calibration on a sample.</strong> Have a person review 1–2% of cases and check the judge agrees;
when they diverge, fix the rubric. Automated eval scales; human eval keeps it honest.</li>
</ul>
<figure class="fig fig-inline"><img src="/blog/assets/media/evaluating-ai-agent-reliability/note-2.png" alt="Hand-drawn sketch from “Evaluating AI Agent Reliability: How to Measure What You Can Trust”" loading="lazy" decoding="async" /></figure>
<h2 id="close-the-loop-with-observability" tabindex="-1">Close the loop with observability <a class="anchor" href="#close-the-loop-with-observability" aria-hidden="true">#</a></h2>
<p>A reliability number is only as good as the data behind it, and a <a href="/#what">multi-agent harness</a> is already
producing most of it. Munder Difflin records <strong>real token and cost telemetry</strong> from the agents’
transcripts, keeps an <a href="/blog/append-only-event-log-agents/">append-only event log</a> of every action, and
tracks task outcomes on a board. That’s an eval dataset waiting to be used: success rate per <em>task type</em>,
cost per <em>successful outcome</em> (a 50x cost spread for similar accuracy is real), and exactly where runs
stall. Feed your <a href="/#how">orchestrator’s</a> own observability back into the eval loop and “is it reliable?”
becomes a trend you watch, not a guess you make.</p>
<h2 id="a-reliability-eval-checklist" tabindex="-1">A reliability-eval checklist <a class="anchor" href="#a-reliability-eval-checklist" aria-hidden="true">#</a></h2>
<ol>
<li><strong>Pick a reliability bar</strong> (and know whether you’re quoting 50% or 80%).</li>
<li><strong>Use benchmarks to rank</strong>, then <strong>re-measure on your own tasks</strong>.</li>
<li><strong>Track the compounding failure modes</strong> — tool use, recovery, context — not just final accuracy.</li>
<li><strong>Ground truth from verifiable checks</strong> (tests); LLM-judge the rest, with rubrics and skepticism.</li>
<li><strong>Run eval in CI</strong> on every change to agent logic; watch regressions.</li>
<li><strong>Calibrate against humans</strong> on a sample, and <strong>feed observability</strong> (cost, outcomes, logs) back in.</li>
</ol>
<p>Do this and “reliable enough to trust” stops being a leap of faith and becomes a measured, improving
number. The agents that earn unattended work are the ones you’ve actually evaluated — on your work, at
your bar.</p>
<p>Want a hive that already records the cost, outcomes, and event logs an eval loop needs? You can
<a href="/#install">download Munder Difflin</a> free — it’s open source.</p>
<section class="faq" aria-label="Frequently asked questions">
<h2 id="faq">FAQ</h2>
<div class="qa">
<p class="q">How do you measure if an AI agent is reliable enough to trust?</p>
<p class="a">Pick a reliability bar and measure against it on tasks like yours. A 50%-reliable agent is a demo; something you'd run unattended needs to be right far more often. METR tracks both 50%- and 80%-reliability horizons for exactly this reason. Then evaluate on your own codebase, not a public benchmark, because curated benchmarks are an optimistic ceiling.</p>
</div>
<div class="qa">
<p class="q">Are benchmarks like SWE-bench enough to judge a coding agent?</p>
<p class="a">No. SWE-bench Verified is a human-curated set of public GitHub issues confirmed to be solvable with fair tests — useful for comparing models, but an optimistic ceiling. Your private, underspecified, weakly-tested codebase is the harder case the benchmark filters out, so real-world reliability is consistently lower than the headline score.</p>
</div>
<div class="qa">
<p class="q">What should an AI agent eval harness measure?</p>
<p class="a">Both correctness and the things that actually break agents over long runs. Use verifiable checks (your test suite) as ground truth for 'did it work,' an LLM-as-judge for fuzzier qualities like readability or security (with care — judges are themselves unreliable), and track tool-use consistency, error recovery, and context management. Run it in CI on every change to agent logic, and calibrate against human review on a small sample.</p>
</div>
</section>
<hr />
<div class="share">
<span class="lbl">Tags</span>
<a class="tag" href="/blog/tags/evaluation/">Evaluation</a><a class="tag" href="/blog/tags/autonomous/">Autonomous</a><a class="tag" href="/blog/tags/agentic-ai/">Agentic AI</a><a class="tag" href="/blog/tags/guardrails/">Guardrails</a>
</div>
</div>
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
<li class="lvl-2"><a href="#reliability-is-a-number-you-have-to-pick">Reliability is a number you have to pick</a></li><li class="lvl-2"><a href="#benchmarks-are-a-ceiling-not-a-verdict">Benchmarks are a ceiling, not a verdict</a></li><li class="lvl-2"><a href="#measure-the-things-that-actually-break">Measure the things that actually break</a></li><li class="lvl-2"><a href="#build-an-eval-harness-on-your-own-work">Build an eval harness on your own work</a></li><li class="lvl-2"><a href="#close-the-loop-with-observability">Close the loop with observability</a></li><li class="lvl-2"><a href="#a-reliability-eval-checklist">A reliability-eval checklist</a></li>
</ol>
</aside>
</div>
<footer class="post-foot wrap"><div class="prevnext">
<a class="pn prev" href="/blog/file-based-coordination-vs-message-queues/"><span class="k">← Older</span><span class="t">File-Based Coordination vs Message Queues for AI Agents</span></a>
<a class="pn next" href="/blog/do-more-with-less-model-routing/"><span class="k">Newer →</span><span class="t">Do More With Less: Smart Model Routing Across Agents</span></a>
</div>
<section class="related">
<h2>Related in Guides</h2>
<div class="post-grid">
<article class="card t-guides">
<a class="thumb t-guides" href="/blog/claude-md-best-practices/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/claude-md-best-practices/hero.png" alt="" loading="lazy" decoding="async" /></a>
<div class="body">
<div class="meta">
<time datetime="2026-10-26">Oct 26, 2026</time>
<span class="dot" aria-hidden="true"></span>
<span>4 min</span>
</div>
<h3><a href="/blog/claude-md-best-practices/">CLAUDE.md Best Practices: What to Write, Where It Lives, How Long (Oct 2026)</a></h3>
<p class="dek">CLAUDE.md best practices from Anthropic's docs, checked 26 Oct 2026: what to put in it, where each file lives, the 200 line target and how AGENTS.md loads.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/claude-md-best-practices/" aria-label="Read: CLAUDE.md Best Practices: What to Write, Where It Lives, How Long (Oct 2026)">Read →</a>
</div>
</div>
</article>
<article class="card t-guides">
<a class="thumb t-guides" href="/blog/codex-computer-use/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/codex-computer-use/hero.png" alt="" loading="lazy" decoding="async" /></a>
<div class="body">
<div class="meta">
<time datetime="2026-10-26">Oct 26, 2026</time>
<span class="dot" aria-hidden="true"></span>
<span>4 min</span>
</div>
<h3><a href="/blog/codex-computer-use/">Codex Computer Use: How to Turn It On, Plans and Limits (Oct 2026)</a></h3>
<p class="dek">Codex computer use lets the ChatGPT desktop app see and click apps on macOS and Windows. Setup, which plans get it, locked use and how Claude's compares.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/codex-computer-use/" aria-label="Read: Codex Computer Use: How to Turn It On, Plans and Limits (Oct 2026)">Read →</a>
</div>
</div>
</article>
<article class="card t-guides">
<a class="thumb t-guides" href="/blog/cursor-pricing/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/cursor-pricing/hero.png" alt="" loading="lazy" decoding="async" /></a>
<div class="body">
<div class="meta">
<time datetime="2026-10-26">Oct 26, 2026</time>
<span class="dot" aria-hidden="true"></span>
<span>5 min</span>
</div>
<h3><a href="/blog/cursor-pricing/">Cursor Pricing (Oct 2026): Every Plan After the 12 Oct Change</a></h3>
<p class="dek">Cursor pricing as of 26 Oct 2026: Free, Plus $8, Premium $30, Super $100, Ultra $200, Teams seats, what changed on 12 Oct and what old plans keep.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/cursor-pricing/" aria-label="Read: Cursor Pricing (Oct 2026): Every Plan After the 12 Oct Change">Read →</a>
</div>
</div>
</article>
</div>
</section>
</footer>
</article>
</main>
<footer class="site-foot">
<div class="wrap">
<div class="foot-top">
<a class="foot-brand brand" href="https://munderdiffl.in" aria-label="Munder Difflin home">
<img class="brand-logo" src="https://munderdiffl.in/logo.png" width="1318" height="840" alt="Munder Difflin, Inc. — Multi-Agent Harness" />
</a>
<div class="foot-links">
<a href="/blog/">Latest</a>
<a href="/blog/topics/">Topics</a>
<a href="/blog/about/">About</a>
<a href="https://munderdiffl.in/blog/feed.xml">RSS</a>
<a href="https://github.com/chaitanyagiri/munder-difflin">GitHub</a>
<a href="https://munderdiffl.in">munderdiffl.in ↗</a>
</div>
</div>
<div class="foot-meta">
© 2026 Munder Difflin · Local-first · MIT-licensed ·
Built on <a href="https://munderdiffl.in">Electron, React, Pixi.js, xterm.js & node-pty</a>.
Run an office of agents while you sleep.
</div>
</div>
</footer>
<script>
// sticky-nav border on scroll
(function () {
var nav = document.getElementById('nav');
if (!nav) return;
var onScroll = function () { nav.classList.toggle('scrolled', window.scrollY > 8); };
onScroll(); window.addEventListener('scroll', onScroll, { passive: true });
})();
// YouTube click-to-load: thumbnail button → youtube-nocookie iframe
(function () {
document.addEventListener('click', function (e) {
var btn = e.target.closest('.yt-load');
if (!btn) return;
var box = btn.closest('.yt');
var id = box && box.getAttribute('data-yt');
// Only an 11 character YouTube video id may reach the iframe URL.
if (!id || !/^[A-Za-z0-9_-]{11}$/.test(id)) return;
var frame = document.createElement('iframe');
frame.src = 'https://www.youtube-nocookie.com/embed/' + encodeURIComponent(id) + '?autoplay=1';
frame.title = box.getAttribute('data-title') || 'YouTube video';
frame.allow = 'accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share';
frame.setAttribute('allowfullscreen', '');
box.replaceChildren(frame);
});
})();
</script>
</body>
</html>