<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Irfan Sadek</title>
    <description>The latest articles on DEV Community by Irfan Sadek (@irfan_sadek_e1222b0dfb7d5).</description>
    <link>https://dev.to/irfan_sadek_e1222b0dfb7d5</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4195579%2F3be0095c-0fef-498a-868b-a60752092653.jpg</url>
      <title>DEV Community: Irfan Sadek</title>
      <link>https://dev.to/irfan_sadek_e1222b0dfb7d5</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/irfan_sadek_e1222b0dfb7d5"/>
    <language>en</language>
    <item>
      <title>Coding AI Training Jobs: What RLHF Code Review Actually Involves (With a Sample Task)</title>
      <dc:creator>Irfan Sadek</dc:creator>
      <pubDate>Mon, 26 Oct 2026 11:44:56 +0000</pubDate>
      <link>https://dev.to/irfan_sadek_e1222b0dfb7d5/coding-ai-training-jobs-what-rlhf-code-review-actually-involves-with-a-sample-task-42j0</link>
      <guid>https://dev.to/irfan_sadek_e1222b0dfb7d5/coding-ai-training-jobs-what-rlhf-code-review-actually-involves-with-a-sample-task-42j0</guid>
      <description>&lt;p&gt;There are &lt;strong&gt;231 open AI-training jobs for software engineers&lt;/strong&gt; on AI Trainer HQ right now, across 10 platforms. The &lt;strong&gt;median listed pay is $65/hr&lt;/strong&gt;, and most roles fall between &lt;strong&gt;$30 and $115/hr&lt;/strong&gt;. Senior roles on 10x Team have a &lt;strong&gt;$139/hr median&lt;/strong&gt;. But "AI coding job" covers very different work, from ranking two Python snippets to debugging an agent across a whole repository. Here's what the work actually involves, which skills pay most, and a realistic sample task.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fyv2yl0smi062r5mk9esd.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fyv2yl0smi062r5mk9esd.png" alt="AI Trainer HQ software engineering jobs page" width="800" height="464"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What RLHF code work involves
&lt;/h2&gt;

&lt;p&gt;RLHF (reinforcement learning from human feedback) for code means &lt;strong&gt;engineers produce the judgments a model learns from.&lt;/strong&gt; Common task types:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Pairwise ranking:&lt;/strong&gt; compare two AI solutions to the same problem, pick the better one and explain why (correctness, edge cases, complexity, readability).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Write the ideal solution:&lt;/strong&gt; produce a reference answer, often with tests.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bug finding:&lt;/strong&gt; identify why AI-generated code fails and write the fix.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prompt writing:&lt;/strong&gt; design hard, realistic coding problems that current models get wrong.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Agentic / repo tasks:&lt;/strong&gt; evaluate an AI agent that edits a real codebase. Does it run, do the tests pass, did it break anything?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Security review:&lt;/strong&gt; flag injection, unsafe deserialisation, secrets in code and similar issues.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;You don't train models directly. You produce the training data.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who's hiring and what they pay
&lt;/h2&gt;

&lt;p&gt;From the &lt;a href="https://aitrainerhq.com/jobs/software-engineering" rel="noopener noreferrer"&gt;software engineering hub&lt;/a&gt;, data as of 26 October 2026:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Platform&lt;/th&gt;
&lt;th&gt;Median&lt;/th&gt;
&lt;th&gt;Range&lt;/th&gt;
&lt;th&gt;Open roles&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;10x Team&lt;/td&gt;
&lt;td&gt;$139/hr&lt;/td&gt;
&lt;td&gt;$65–194/hr&lt;/td&gt;
&lt;td&gt;10&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Welo Global&lt;/td&gt;
&lt;td&gt;$105/hr&lt;/td&gt;
&lt;td&gt;$105/hr&lt;/td&gt;
&lt;td&gt;11&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Alignerr&lt;/td&gt;
&lt;td&gt;$83/hr&lt;/td&gt;
&lt;td&gt;$60–100/hr&lt;/td&gt;
&lt;td&gt;2&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Mercor&lt;/td&gt;
&lt;td&gt;$70/hr&lt;/td&gt;
&lt;td&gt;$17–500/hr&lt;/td&gt;
&lt;td&gt;52&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Meridial&lt;/td&gt;
&lt;td&gt;$65/hr&lt;/td&gt;
&lt;td&gt;$6–150/hr&lt;/td&gt;
&lt;td&gt;65&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Anyone AI&lt;/td&gt;
&lt;td&gt;$63/hr&lt;/td&gt;
&lt;td&gt;$15–80/hr&lt;/td&gt;
&lt;td&gt;16&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;micro1&lt;/td&gt;
&lt;td&gt;$60/hr&lt;/td&gt;
&lt;td&gt;$6–350/hr&lt;/td&gt;
&lt;td&gt;47&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Rise Data Labs&lt;/td&gt;
&lt;td&gt;$30/hr&lt;/td&gt;
&lt;td&gt;$15–50/hr&lt;/td&gt;
&lt;td&gt;17&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Mercor has listed US software engineers at $70–150/hr this month. Several current Mercor roles require you to &lt;strong&gt;own a MacBook&lt;/strong&gt; ($50–60/hr) for macOS app-usage projects. Anyone AI hires developers by country, mostly in Latin America.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; AI Trainer HQ earns a referral fee only from micro1. That doesn't affect the data. See &lt;a href="https://aitrainerhq.com/disclosure" rel="noopener noreferrer"&gt;/disclosure&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h3&gt;
  
  
  Country access
&lt;/h3&gt;

&lt;p&gt;Coding roles open by country: &lt;strong&gt;US 123, India 94, UK 90, Philippines 80, Kenya 78.&lt;/strong&gt; Coding is one of the most globally open categories, so check yours on the &lt;a href="https://aitrainerhq.com/tools/eligibility" rel="noopener noreferrer"&gt;eligibility tool&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which languages and skills pay most
&lt;/h2&gt;

&lt;p&gt;Listings don't break pay down by programming language, but patterns are clear from the roles:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Python&lt;/strong&gt; is the default almost everywhere, so it's table stakes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;TypeScript/JavaScript, Java, Go, Rust and C++&lt;/strong&gt; appear in specialist roles. Rust and C++ often sit in higher-paid systems work.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Infra and ML engineering&lt;/strong&gt; (kernels, profiling, LLM serving): one current Mercor MLOps role lists $90–120/hr.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Language plus coding:&lt;/strong&gt; Meridial lists "Coding Specialist (Fluent in Japanese / Bengali / Dutch / Swedish / Chinese / Italian / German)" at $65/hr.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Seniority beats language.&lt;/strong&gt; 10x Team's $139/hr median is driven by senior engineers and CTO-level reviewers.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  A sample task walkthrough
&lt;/h2&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Prompt:&lt;/strong&gt; "Write a Python function that returns the k most frequent words in a string, case-insensitive, ties broken alphabetically."&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Response A:&lt;/strong&gt;&lt;/p&gt;


&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;top_k&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;s&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;k&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;collections&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;Counter&lt;/span&gt;
    &lt;span class="n"&gt;c&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;Counter&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;s&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;lower&lt;/span&gt;&lt;span class="p"&gt;().&lt;/span&gt;&lt;span class="nf"&gt;split&lt;/span&gt;&lt;span class="p"&gt;())&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;w&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;w&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;_&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;c&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;most_common&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;k&lt;/span&gt;&lt;span class="p"&gt;)]&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;


&lt;p&gt;&lt;strong&gt;Response B:&lt;/strong&gt;&lt;/p&gt;


&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;re&lt;/span&gt;
&lt;span class="kn"&gt;from&lt;/span&gt; &lt;span class="n"&gt;collections&lt;/span&gt; &lt;span class="kn"&gt;import&lt;/span&gt; &lt;span class="n"&gt;Counter&lt;/span&gt;
&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;top_k&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;s&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;k&lt;/span&gt;&lt;span class="p"&gt;):&lt;/span&gt;
    &lt;span class="n"&gt;words&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;re&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;findall&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sa"&gt;r&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;[a-z&lt;/span&gt;&lt;span class="sh"&gt;'&lt;/span&gt;&lt;span class="s"&gt;]+&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;s&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;lower&lt;/span&gt;&lt;span class="p"&gt;())&lt;/span&gt;
    &lt;span class="n"&gt;c&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;Counter&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;words&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;sorted&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;c&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;key&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="k"&gt;lambda&lt;/span&gt; &lt;span class="n"&gt;w&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="n"&gt;c&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;w&lt;/span&gt;&lt;span class="p"&gt;],&lt;/span&gt; &lt;span class="n"&gt;w&lt;/span&gt;&lt;span class="p"&gt;))[:&lt;/span&gt;&lt;span class="n"&gt;k&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/blockquote&gt;

&lt;p&gt;&lt;strong&gt;A strong review:&lt;/strong&gt;&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Correctness:&lt;/strong&gt; A ignores the tie-break rule (&lt;code&gt;most_common&lt;/code&gt; keeps insertion order on ties) and treats "word," and "word" as different words. B handles both.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Edge cases:&lt;/strong&gt; B returns fewer than k items if there aren't enough words, which is fine. Note that neither validates &lt;code&gt;k &amp;lt;= 0&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Complexity:&lt;/strong&gt; B sorts all unique words, O(n log n), which is acceptable. Mention &lt;code&gt;heapq.nsmallest&lt;/code&gt; for very large inputs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Verdict:&lt;/strong&gt; B is better. Give a confidence level and a one-paragraph justification that cites the spec.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ideal answer&lt;/strong&gt; (if requested): B plus input validation and a docstring with an example.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That's the skill being paid for: &lt;strong&gt;precise, spec-grounded reasoning&lt;/strong&gt;, written clearly.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Screen recording (GIF attached: /workspace/aitrainerhq/recordings/aitrainerhq-browse-and-filter-jobs.gif):&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff3iy3d96xqbz3pi7prgy.gif" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ff3iy3d96xqbz3pi7prgy.gif" alt="screen recording" width="720" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  How to pass a coding assessment
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Run the code.&lt;/strong&gt; Many reviewers skip it and miss obvious failures.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Quote the spec&lt;/strong&gt; in your justification.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Don't over-engineer the ideal answer.&lt;/strong&gt; Graders want correct, idiomatic and readable code.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Never paste AI output&lt;/strong&gt; as your answer. Platforms check for it.&lt;/li&gt;
&lt;li&gt;Prepare a short portfolio: a GitHub profile with real work helps a lot on Mercor and 10x Team.&lt;/li&gt;
&lt;li&gt;Full guide: &lt;a href="https://aitrainerhq.com/guides" rel="noopener noreferrer"&gt;this guide&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Tools and setup that help
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A local environment&lt;/strong&gt; with Python 3.x, Node and Docker, so you can actually run the code in tasks&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A linter and formatter&lt;/strong&gt; (ruff, eslint) to spot style issues quickly&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A scratch test file&lt;/strong&gt; for quick property checks on edge cases&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A reliable second monitor.&lt;/strong&gt; Long side-by-side reviews are much faster with one.&lt;/li&gt;
&lt;li&gt;For agentic tasks, comfort with &lt;strong&gt;git diffs&lt;/strong&gt;. You'll often be reviewing a patch, not a single function.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Realistic earnings
&lt;/h2&gt;

&lt;p&gt;At the $65/hr median, 10 hours a week is about &lt;strong&gt;$650 before tax&lt;/strong&gt;. Senior engineers on 10x Team can do much better, but those roles have strict screening and fewer seats (10 open right now).&lt;/p&gt;

&lt;h2&gt;
  
  
  Caveats
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Advertised ≠ earned.&lt;/strong&gt; Task-based coding work can drop your effective hourly rate if tasks run long.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Queues dry up.&lt;/strong&gt; Coding projects end when a model release ships. Keep 2–3 platforms going.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Confidentiality:&lt;/strong&gt; you'll often work on code for unreleased models, so don't post tasks publicly.&lt;/li&gt;
&lt;li&gt;A degree isn't always required. Many listings accept equivalent experience, proven through the coding test.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How we got these numbers
&lt;/h2&gt;

&lt;p&gt;Every figure here comes from AI Trainer HQ's own tracking. We pull live jobs only from each platform's own applicant tracking system, re-check them at the source every 24 hours, and convert pay to US dollars per hour. Medians use the midpoint of each role's listed range, and roles that don't state pay are left out. The full method is on our &lt;a href="https://aitrainerhq.com/methodology" rel="noopener noreferrer"&gt;methodology page&lt;/a&gt;. Numbers move daily, so treat this post as a dated snapshot and confirm the details on the live listing before you apply.&lt;/p&gt;

&lt;h2&gt;
  
  
  Next step
&lt;/h2&gt;

&lt;p&gt;Open the &lt;a href="https://aitrainerhq.com/jobs/software-engineering" rel="noopener noreferrer"&gt;software engineering hub&lt;/a&gt;, filter by country, and apply to one high-volume platform (Mercor or Meridial) and one senior-focused one (10x Team) if your experience fits.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>career</category>
      <category>programming</category>
      <category>remotework</category>
    </item>
    <item>
      <title>Mercor vs micro1 vs Mindrift for Expert (SME) AI Training Work in 2026</title>
      <dc:creator>Irfan Sadek</dc:creator>
      <pubDate>Mon, 26 Oct 2026 11:34:16 +0000</pubDate>
      <link>https://dev.to/irfan_sadek_e1222b0dfb7d5/mercor-vs-micro1-vs-mindrift-for-expert-sme-ai-training-work-in-2026-3f5h</link>
      <guid>https://dev.to/irfan_sadek_e1222b0dfb7d5/mercor-vs-micro1-vs-mindrift-for-expert-sme-ai-training-work-in-2026-3f5h</guid>
      <description>&lt;p&gt;If you're a doctor, lawyer, engineer or PhD looking at expert AI-training work, three names come up constantly. Here's how they compare on advertised pay today: &lt;strong&gt;Mercor has a median of $80/hr across 297 open roles, Mindrift $69/hr across 24, and micro1 $65/hr across 294.&lt;/strong&gt; But the median is only part of the story. Onboarding, payout schedules, country rules and how long postings stay open matter as much as the hourly number.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Disclosure:&lt;/strong&gt; AI Trainer HQ is free for workers. We earn a referral fee only from &lt;strong&gt;micro1&lt;/strong&gt; today. We earn nothing from Mercor or Mindrift. That's why we stick to published data below, and you can check every number on the &lt;a href="https://aitrainerhq.com/pay-index" rel="noopener noreferrer"&gt;Pay Index&lt;/a&gt; and our &lt;a href="https://aitrainerhq.com/disclosure" rel="noopener noreferrer"&gt;disclosure page&lt;/a&gt;.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkaegx4xq62qmeyh2p8lr.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fkaegx4xq62qmeyh2p8lr.png" alt="Pay Index ranking of platforms by median pay" width="800" height="464"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The comparison at a glance
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Mercor&lt;/th&gt;
&lt;th&gt;micro1&lt;/th&gt;
&lt;th&gt;Mindrift&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Open roles (26 Oct 2026)&lt;/td&gt;
&lt;td&gt;297&lt;/td&gt;
&lt;td&gt;294&lt;/td&gt;
&lt;td&gt;24&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Median advertised pay&lt;/td&gt;
&lt;td&gt;$80/hr&lt;/td&gt;
&lt;td&gt;$65/hr&lt;/td&gt;
&lt;td&gt;$69/hr&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Advertised range&lt;/td&gt;
&lt;td&gt;$15–500/hr&lt;/td&gt;
&lt;td&gt;$2–400/hr&lt;/td&gt;
&lt;td&gt;$9–100/hr&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Roles stating pay&lt;/td&gt;
&lt;td&gt;296 of 297&lt;/td&gt;
&lt;td&gt;288 of 294&lt;/td&gt;
&lt;td&gt;19 of 24&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Median days a role is listed&lt;/td&gt;
&lt;td&gt;42&lt;/td&gt;
&lt;td&gt;19&lt;/td&gt;
&lt;td&gt;7&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Roles listed 60+ days&lt;/td&gt;
&lt;td&gt;28%&lt;/td&gt;
&lt;td&gt;4%&lt;/td&gt;
&lt;td&gt;0%&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Screening&lt;/td&gt;
&lt;td&gt;Résumé, ~20-min AI interview, assessments&lt;/td&gt;
&lt;td&gt;AI interviewer "Zara," then project tests&lt;/td&gt;
&lt;td&gt;CV review, assessments, ID check&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Countries&lt;/td&gt;
&lt;td&gt;Most countries&lt;/td&gt;
&lt;td&gt;Most countries&lt;/td&gt;
&lt;td&gt;Most, but not China, Russia, Iran, Kenya, Switzerland or Taiwan&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Payout&lt;/td&gt;
&lt;td&gt;Stripe (bank), Wise&lt;/td&gt;
&lt;td&gt;Varies by contract&lt;/td&gt;
&lt;td&gt;Payoneer or PayPal via Tipalti, twice monthly (5th and 20th)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Referral fee to AI Trainer HQ&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;td&gt;&lt;strong&gt;Yes&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;For the full picture, read each review: &lt;a href="https://aitrainerhq.com/platforms/mercor" rel="noopener noreferrer"&gt;Mercor&lt;/a&gt;, &lt;a href="https://aitrainerhq.com/platforms/micro1" rel="noopener noreferrer"&gt;micro1&lt;/a&gt;, &lt;a href="https://aitrainerhq.com/platforms/mindrift" rel="noopener noreferrer"&gt;Mindrift&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Pay: what the numbers actually say
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Mercor&lt;/strong&gt; has the highest median of the three. It also has the widest spread, because it lists both $30/hr generalist feedback studies and $350/hr radiologist roles. Physician roles we track on Mercor commonly list $150–350/hr. Legal roles run $100–150/hr for JD/LLB holders with a current licence.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;micro1&lt;/strong&gt; has a similar median, but its top end is very high for specialist attorneys. Several current roles list $140–400/hr, and its legal median is $125/hr. Its bottom end ($2/hr) comes from a small number of task-based or regional roles, so read the specific listing.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Mindrift&lt;/strong&gt; is narrower. Most current roles are physician evaluations at $35–100/hr and engineering (CAD/CFD) roles at around $69/hr. It doesn't reach Mercor's peaks, but it shows a fixed reward on each task before you start, so you know what a task pays up front.&lt;/p&gt;

&lt;p&gt;For context: 10x Team, which isn't one of the three, has the highest median we track at $135/hr. It hires senior specialists only.&lt;/p&gt;

&lt;h2&gt;
  
  
  Onboarding: how hard is it to get in?
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Mercor:&lt;/strong&gt; a résumé screen, then a ~20-minute AI interview, then domain assessments. The interview is the main filter. Many people pass and then wait in a talent pool for a matching project.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;micro1:&lt;/strong&gt; the AI interviewer Zara does an early screen, followed by project-specific tests. Postings turn over fast (median 19 days listed), which suggests real, current projects rather than evergreen pools.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mindrift:&lt;/strong&gt; apply with a CV, take assessments and verify your ID. Its "all-access" tasks need no application at all, which makes it the easiest of the three to start on, though those tasks pay less.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Our guide to passing these screens is here: &lt;a href="https://aitrainerhq.com/guides" rel="noopener noreferrer"&gt;this guide&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Screen recording:&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7y7wolb4kp0cri2dqbzv.gif" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F7y7wolb4kp0cri2dqbzv.gif" alt="screen recording" width="720" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Project stability: the part nobody advertises
&lt;/h2&gt;

&lt;p&gt;The "median days listed" figure is a useful signal:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;A role listed for &lt;strong&gt;months&lt;/strong&gt; is usually an &lt;strong&gt;evergreen talent pool&lt;/strong&gt;. The platform keeps the posting open and contacts you when a project starts. 28% of Mercor's roles are over 60 days old.&lt;/li&gt;
&lt;li&gt;A role listed for &lt;strong&gt;days&lt;/strong&gt; usually means a &lt;strong&gt;live project&lt;/strong&gt; hiring now. Mindrift's median is 7 days, and micro1's is 19.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Neither is better by default. Evergreen pools mean you might wait weeks after acceptance. Fresh postings mean the project may also end sooner. Mindrift's own review notes that workers report weeks-long waits after onboarding, and its Trustpilot rating is 3.4/5.&lt;/p&gt;

&lt;h2&gt;
  
  
  Payout reliability
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Mindrift&lt;/strong&gt; pays on a fixed schedule: the 5th and 20th, for work accepted in the previous half-month, through Payoneer or Tipalti/PayPal, with no commission. Toloka runs it, and Toloka raised $72M in May 2025 in a round led by Bezos Expeditions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mercor&lt;/strong&gt; pays through Stripe bank transfer or Wise. Check that Stripe supports your country.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;micro1&lt;/strong&gt; payout terms depend on your contractor agreement. Read it before you start.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Our Legit Score weights payout reliability at 35%. We publish a score only once we've checked the evidence, and none of the three is rated yet, so we're not going to pretend otherwise.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which one should you pick?
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Licensed physician or specialist:&lt;/strong&gt; apply to &lt;strong&gt;Mercor and Mindrift&lt;/strong&gt;. Both have deep medicine pipelines.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Attorney:&lt;/strong&gt; &lt;strong&gt;micro1 and Mercor&lt;/strong&gt;. micro1's specialist attorney roles have the highest ceilings.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Engineer (CAD/CFD/mechanical):&lt;/strong&gt; &lt;strong&gt;Mindrift&lt;/strong&gt; currently has the clearest engineering roles. Mercor has MacBook-required engineering roles at $50–60/hr.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Based in Kenya:&lt;/strong&gt; skip Mindrift (unsupported). Mercor and Meridial are your main options.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Want to start this week with no application:&lt;/strong&gt; Mindrift's all-access tasks.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Honestly, the best answer for most experts is &lt;strong&gt;apply to two&lt;/strong&gt;. Queues empty without warning, and a second platform smooths your income.&lt;/p&gt;

&lt;h2&gt;
  
  
  A simple scoring sheet for your own decision
&lt;/h2&gt;

&lt;p&gt;Before you apply, rate each platform 1–5 on these five questions:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;How many live roles are there &lt;strong&gt;in my exact field&lt;/strong&gt;?&lt;/li&gt;
&lt;li&gt;Is the median for my field above my minimum acceptable rate?&lt;/li&gt;
&lt;li&gt;Can its payout method actually pay me where I live?&lt;/li&gt;
&lt;li&gt;Are its roles fresh (live projects) or evergreen (waiting pools)?&lt;/li&gt;
&lt;li&gt;How long is the screening, and is any of it unpaid?&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Apply first to whichever scores highest. It's a quick exercise, but it keeps you from chasing the top-of-range number.&lt;/p&gt;

&lt;h2&gt;
  
  
  Caveats
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;These are &lt;strong&gt;advertised&lt;/strong&gt; rates, not reported earnings. Effective hourly pay can be lower on task-based work.&lt;/li&gt;
&lt;li&gt;Counts and medians change daily. Always check the live listing.&lt;/li&gt;
&lt;li&gt;We have 0 verified worker reports on Mindrift so far, and we only publish payout timing once 10 have come in.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Can I work on all three at once?&lt;/strong&gt; Usually, yes, since they're independent contractor arrangements. Check each contract for exclusivity or confidentiality clauses, especially on projects for the same AI lab.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Which one pays fastest?&lt;/strong&gt; Mindrift is the only one of the three with a published fixed schedule (twice monthly). For the others, check your contract terms.&lt;/p&gt;

&lt;h2&gt;
  
  
  Next step
&lt;/h2&gt;

&lt;p&gt;Open the &lt;a href="https://aitrainerhq.com/pay-index" rel="noopener noreferrer"&gt;Pay Index&lt;/a&gt; and filter by your profession, then apply to the two platforms with the most live roles in your field. You can download the CSV if you want to dig in yourself.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>career</category>
      <category>productivity</category>
      <category>discuss</category>
    </item>
    <item>
      <title>Auto-Caption and Schedule Shorts From Claude Code: A Terminal-First Workflow</title>
      <dc:creator>Irfan Sadek</dc:creator>
      <pubDate>Tue, 20 Oct 2026 07:22:57 +0000</pubDate>
      <link>https://dev.to/irfan_sadek_e1222b0dfb7d5/auto-caption-and-schedule-shorts-from-claude-code-a-terminal-first-workflow-294c</link>
      <guid>https://dev.to/irfan_sadek_e1222b0dfb7d5/auto-caption-and-schedule-shorts-from-claude-code-a-terminal-first-workflow-294c</guid>
      <description>&lt;p&gt;Developers make a lot of video without meaning to: release demos, conference talks, pairing sessions, livestreams, Loom walkthroughs. Most of it ends up sitting in a &lt;code&gt;recordings/&lt;/code&gt; folder, because turning a 40-minute talk into five captioned shorts means opening an editor, scrubbing the timeline, exporting, captioning, uploading and scheduling. Nobody with a backlog of pull requests does that.&lt;/p&gt;

&lt;p&gt;But you already have an agent in your terminal. Claude Code can call remote tools over MCP, so the whole "clip, caption, schedule" pipeline can be a single prompt run from your project directory. This post walks through the setup, the prompt patterns that work, and the parts to keep a human in the loop for.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why do this from Claude Code at all?
&lt;/h2&gt;

&lt;p&gt;A browser clipping tool works fine. The terminal has three advantages for developers:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Context.&lt;/strong&gt; Claude Code can read your changelog, README and release notes, so it writes clip titles and captions that are accurate about what shipped.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Batching.&lt;/strong&gt; Point it at a folder and it works through every file, without you uploading them one by one.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Repeatability.&lt;/strong&gt; Save the prompt in your repo (say in &lt;code&gt;CLAUDE.md&lt;/code&gt; or a slash command) and every release gets the same treatment.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Step 1: Add a remote MCP server
&lt;/h2&gt;

&lt;p&gt;You need a server that exposes clipping, captioning and scheduling tools. I'll use &lt;a href="https://tryunsora.com" rel="noopener noreferrer"&gt;Unsora&lt;/a&gt;, which exposes those (plus video generation) over remote MCP with Streamable HTTP:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;claude mcp add &lt;span class="nt"&gt;--transport&lt;/span&gt; http unsora https://mcp.tryunsora.com/mcp
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then check it's connected:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;claude mcp list
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The first time you call a tool, it asks you to sign in through the browser. Your credits and connected social accounts stay with that login. Setup for Cursor and other clients is on the &lt;a href="https://tryunsora.com/connector" rel="noopener noreferrer"&gt;connector page&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If your team shares the project, you can commit the server entry to &lt;code&gt;.mcp.json&lt;/code&gt; so everyone gets it. Just never commit API keys. If you use the REST API instead of OAuth, put the key in an environment variable.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: Understand the tool chain
&lt;/h2&gt;

&lt;p&gt;Once it's connected, ask Claude Code to list the server's tools. The ones that matter for this workflow follow a simple async pattern:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Step&lt;/th&gt;
&lt;th&gt;Tool&lt;/th&gt;
&lt;th&gt;What it does&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Clip&lt;/td&gt;
&lt;td&gt;&lt;code&gt;create_clipping&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Finds the strongest moments, reframes to 9:16 with the speaker centred, burns in word-by-word captions&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Wait&lt;/td&gt;
&lt;td&gt;&lt;code&gt;wait_for_clipping&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Polls until the job finishes (minutes, not hours)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Thumbnail&lt;/td&gt;
&lt;td&gt;&lt;code&gt;create_thumbnail&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Optional cover image per clip&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Schedule&lt;/td&gt;
&lt;td&gt;&lt;code&gt;create_post&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;Queues the post to YouTube, TikTok, Instagram, Facebook or LinkedIn&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Clipping is asynchronous, so the agent has to submit a job and then wait for it. That's exactly what agents handle well and shell scripts handle poorly.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: The one-prompt pipeline
&lt;/h2&gt;

&lt;p&gt;From your repo root:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Take recordings/v2.4-launch-demo.mp4. Find the 5 strongest moments about
new features (use CHANGELOG.md to know what's new). Cut them 9:16 with
yellow captions. Give each a hook title under 60 characters. Then schedule
one per day to YouTube Shorts and LinkedIn starting Monday 9:00, with a
different caption per platform. Show me the plan before scheduling.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;What happens:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Claude Code reads &lt;code&gt;CHANGELOG.md&lt;/code&gt; to learn what shipped.&lt;/li&gt;
&lt;li&gt;It calls &lt;code&gt;create_clipping&lt;/code&gt; with your instructions ("moments about new features").&lt;/li&gt;
&lt;li&gt;It waits for the job, then shows you the clips with timestamps.&lt;/li&gt;
&lt;li&gt;It drafts captions per platform. LinkedIn gets a more professional tone and Shorts gets a hook.&lt;/li&gt;
&lt;li&gt;It stops, because you asked to see the plan first, and only calls &lt;code&gt;create_post&lt;/code&gt; once you approve.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That last line matters. Nothing should post without a human saying yes. Unsora also won't publish until you confirm, but it's good practice to put the checkpoint in the prompt too.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: Batch a whole folder
&lt;/h2&gt;

&lt;p&gt;The real win is the backlog:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;For every .mp4 in recordings/talks/, pull the 3 best moments, caption them
in our brand style (white text, black outline), and queue them across the
next three weeks on TikTok and YouTube, max one post per day per platform.
Write a summary table to clips-plan.md before scheduling anything.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Writing the plan to a file gives you something to review in a diff, which suits developers better than scrolling a chat log.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: Make it a reusable command
&lt;/h2&gt;

&lt;p&gt;Save the prompt as a project slash command, for example &lt;code&gt;.claude/commands/ship-shorts.md&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight markdown"&gt;&lt;code&gt;Clip $ARGUMENTS into 3-5 captioned 9:16 shorts focused on what's new in
CHANGELOG.md. Hook titles under 60 chars. Draft per-platform captions.
Write the plan to clips-plan.md and wait for my approval before scheduling.
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Now every release ends with &lt;code&gt;/ship-shorts recordings/latest.mp4&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prompt tips that improve output
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Describe the moments by meaning.&lt;/strong&gt; "Every time I run the CLI" or "the part where I explain the migration" works better than timestamps.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Name the caption style once and reuse it.&lt;/strong&gt; Consistency across clips reads as a brand.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Ask for more clips than you need.&lt;/strong&gt; Picking 5 out of 10 is faster than re-running.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Translate captions&lt;/strong&gt; if your users aren't mainly English speakers. Ask for it per clip.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Need fresh b-roll for a gap?&lt;/strong&gt; The same server can generate it with Seedance, Kling or Veo models. See the &lt;a href="https://tryunsora.com/ai-video-generator" rel="noopener noreferrer"&gt;AI video generator page&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What to keep human
&lt;/h2&gt;

&lt;p&gt;Agents are good at the mechanics, not the judgement. Review:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Accuracy.&lt;/strong&gt; Does the clip claim a feature works in a way it doesn't?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Context loss.&lt;/strong&gt; A 30-second cut of a nuanced point can mislead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Platform fit.&lt;/strong&gt; Some clips belong on LinkedIn only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Posting cadence.&lt;/strong&gt; Don't flood a channel just because the agent can.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Cost and limits
&lt;/h2&gt;

&lt;p&gt;Clip runs take a few minutes, longer for a two-hour source. Unsora is paid (from $19/month with a 3-day trial). The MCP server, web app and REST API all use the same account and credits, so you can move between terminal and browser. Frame-accurate editing still belongs in a real editor. This workflow is for short-form repurposing.&lt;/p&gt;

&lt;h2&gt;
  
  
  Wrapping up
&lt;/h2&gt;

&lt;p&gt;If you've already connected a video MCP server to Claude, this is the natural next step. I covered the basics of &lt;a href="https://sadekxd.github.io/claude-custom-connector-video-generation/" rel="noopener noreferrer"&gt;adding a custom connector for video generation&lt;/a&gt; in an earlier guide. Clipping and scheduling turn your recordings folder from storage into a content pipeline you can run from the same terminal you ship code from.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>claude</category>
      <category>mcp</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
