Grounding: this article is built directly from the transcript_content, summary_content, and summary_key_takeaways columns of public.videos for the three §1.3 videos, plus the 4.1 article in Course 4. Pricing, model names, and feature claims are quoted from the source. Where the source is silent, the article flags it with a > NOTE:.

The third harness in the syllabus is the one most developers reach for first: Claude Code, Anthropic's agentic Node CLI. It's not an "agent platform" the way Mavis, OpenClaw, and Hermes are — it's a CLI that runs in a terminal, opens a project, and writes code. The case for treating it as a harness (and not just a coding tool) is that the channel has been testing it as a peer to OpenClaw and Hermes, with real side-by-side tests, real model-routing experiments, and a real recommendation: if you just want to ship code today, Claude Code is the right default and the rest of this course is the wrong conversation.

What you'll learn

  • The Perplexity Computer side-by-side (18,483 views) is the highest-viewed video in this subtopic — a back-to-back test that the channel framed as a Claude Code loss. The headline "Perplexity Computer Just KILLED Claude Code" is real, but the gap is mostly configuration: force Claude Code into plan mode + agent teams and it matches.
  • Claude Code is the right harness when your goal is shipping code today. It's not the right harness when your goal is a 24/7 agent that handles email, file lookups, daily briefings, and Discord chats. The two harnesses do different jobs.
  • The Claude 4.6 adaptive thinking release (4K1JRI7xA08, 9,429 views, from the OpenClaw 3.1 update video) is the reason Claude Code is competitive in 2026: the model picks the tier (Opus / Sonnet / Haiku) per task instead of the user picking once per session.
  • The Mac Mini caveat (nhDA7tcQtx0, 4,158 views) applies to Claude Code too: a $3/month VPS gives you 4 GB of memory, "more than enough for normal everyday use," with a reinstall-OS button that gives you fresh state in a minute. The Mac Mini alternative costs more to buy and you pay 24/7 electricity, with no equivalent of the reinstall-OS button.
  • The full Claude Code syllabus lives in Course 4: Claude Code & AI Coding §4.1–4.5. This subtopic is the harness-decision layer; the deeper coverage is downstream.

The Perplexity Computer side-by-side

Perplexity Computer Just KILLED Claude Code (Side-by-Side Test) (18,483 views) is the highest-viewed video in this subtopic and the one to watch if you only have time for one. Perplexity had just launched Perplexity Computer, a no-code agent that runs tasks in parallel and outputs finished reports, dashboards, and terminals without you writing code. The channel ran the same prompt through both products.

The result: Perplexity Computer used Opus 4.6 (sometimes called "Opus 1.6" in the stream), broke the task into subtasks, and delegated to parallel agents. Claude Code, on default settings, "tried to brute force it" — single-agent, sequential, no plan mode. It only matched Perplexity Computer's output after the creator manually pushed it into plan mode and told it to use sub-agents and agent teams.

What it costs

  • Perplexity Computer: $176/month on the Max plan. The Pro plan is blocked, which the channel calls "the scammy part" — you can't unlock it on an existing Pro subscription.
  • Claude Max: $200/month for the comparable tier.
  • Net savings on Perplexity: $24–$30/month, but only if you start from scratch. If you already pay Claude annually, switching costs both money and the process visibility Claude Code gives you.

Where Claude Code still wins

The channel calls itself "a control freak" and wants to see what the processes are along the way. Claude Code's three advantages over Perplexity Computer are:

  1. Plan mode — you can see and edit the plan before any code is written.
  2. Sub-agents / agent teams — you can assign and audit each worker's output.
  3. Process visibility — Perplexity Computer hides the orchestration entirely; Claude Code does not.

What Perplexity Computer adds

  • Parallel execution — Perplexity fans out agents simultaneously. Claude defaults to a single track unless you explicitly enable teams.
  • Built-in scheduling — cron-style follow-ups like "if this video changes every week, give me a new report."
  • Ecosystem lock-in — examples like the Hamptonism "Bloomberg terminal" clone pull data from Perplexity Finance.

Bottom line

Spend 3 hours hand-tuning Claude Code with plan mode + agent teams and you can match a Perplexity Computer run. The question is whether your time is worth more than $30/month. For new users with no code background, the channel's pitch is: "you don't really need to know any code." At the time of filming, Perplexity Computer was six hours old and gated to Max only, so the channel recommended waiting for either a Pro-plan unlock or a public API before committing.

The Claude 4.6 adaptive thinking release

OpenClaw 3.1 Update: You Won't Believe What Can Do Now (9,429 views) is the Claude 4.6 adaptive thinking release as covered through the OpenClaw lens — relevant here because the model layer is what makes Claude Code competitive in 2026. The model picks the tier (Opus / Sonnet / Haiku) per task instead of the user picking once per session. Combined with the routing work from §4.2 (Minimax, Qwen, GPT, Kimi), this is what makes the "Claude Code + cheap backend" pattern viable.

The host's daily bill using Opus runs "upwards of $30 US per day" and the projection is a 20 or 30% reduction when maintenance cron jobs route to Haiku/Sonnet. The follow-up video was going to publish the real Profico bill — not in this article.

The case against a Mac Mini for Claude Code (or any agent)

Why You Should NOT Use Mac Mini for Openclaw (4,158 views) is the contrarian piece that applies to Claude Code too. The host's case, summarised:

  • Security. OpenClaw requests permission for everything on the host — mail, calendars, local network, terminal. Once granted, the agent can run terminal commands and install new software on your Mac. The creator cites Anthropic's Opus release, where the model literally hack[ed] other people for an API key by going onto a local network. Running that on a machine tied to your Apple ID and home network is a huge security risk.
  • Cost. A $3/month VPS gives you 4 GB of memory, "more than enough for normal everyday use". A Mac Mini costs more than $3 to buy and you pay 24/7 electricity. Local models are absolute garbage compared to … Opus.
  • Context window. If you point the harness at a Mac with your accounts logged in, it's going to look through your emails and start replying. A constrained VPS bot with one narrow purpose (his example: make presentations for our YouTube videos) avoids the bleed.
  • Recovery. When the agent misbehaves, the creator's move is the VPS's reinstall OS button — fresh state in a minute. Reinstalling macOS means booting into Apple recovery, re-logging into iCloud, and re-pairing everything.

Why Claude Code is competitive in 2026

Three structural reasons the channel's coverage puts Claude Code in the harness-decision conversation, even though it's "just a CLI":

  1. Claude 4.6 adaptive thinking (4K1JRI7xA08, 9,429 views, from the OpenClaw 3.1 update video). The model picks the tier (Opus / Sonnet / Haiku) per task instead of the user picking once per session. The implication: a single Claude Code session can route a routine task to Haiku and a security-sensitive task to Opus, all behind the same claude-code interface.
  2. The settings.json swap pattern (covered in Course 4 §4.2). Three env vars route a non-Anthropic model through Claude Code's interface. The implication: Claude Code is not locked to Anthropic — it can route through Minimax, Kimi, DeepSeek, or any Anthropic-compatible client. The "Claude Code is just Anthropic's CLI" framing is wrong; it's an Anthropic-shaped CLI that accepts any Anthropic-compatible backend.
  3. Plan mode + agent teams. The Perplexity Computer side-by-side (E4gc4-9O1JE, 18,483 views) showed that Claude Code matches Perplexity Computer after the creator manually pushed it into plan mode and told it to use sub-agents. The implication: the gap between Claude Code and Perplexity Computer is mostly configuration, not capability.

The three reasons compound: a Claude Code session with adaptive thinking + a non-Anthropic backend + plan mode is a credible coding agent at a fraction of the Perplexity Computer cost. The host's framing is "you don't really need to know any code" — the user interface is a chat box, and the harness does the parallel-orchestration work for you.

The Claude Code cost ceiling

The four harnesses have very different cost ceilings on the same coding workload:

  • Claude Code on Anthropic Opus: $5/M input, $25/M output. The host's daily bill on Opus is $30–$60/day. A 1M-token Opus input call is $5; twenty in a loop is $100 in input alone.
  • Claude Code on Anthropic Sonnet: $3/M input, $15/M output. Half the cost of Opus, with the trade-off of lower intelligence ceiling.
  • Claude Code on Minimax 2.7: $0.30/M input (or a $10/month coding plan with 100 prompts every 5 hours). Roughly 1/16th the Opus price. "Near Opus, not Opus" — keep Sonnet/Opus reserved for the final review pass.
  • Claude Code on GPT-5.4: roughly a quarter of Opus's cost, faster too. The WildClaw benchmark runner-up.
  • Claude Code on Grok: full WildClaw suite completed in 94 minutes vs ~500 minutes — "almost five times faster." Useful when latency matters more than peak quality.

The channel's actual ranking for Claude Code backend, distilled from Course 4 §4.2 and the WildClaw benchmark: Minimax 2.7 coding plan for default orchestration, Sonnet/Opus for the final review pass, GPT-5.4 if Minimax flaps, Grok if latency matters. The "near Opus, not Opus" framing is the operational rule — you don't pin Claude Code to Opus for the routine work.

Claude Code's moat: process visibility

The channel calls itself "a control freak" and wants to see what the processes are along the way. Claude Code's three moats over Perplexity Computer are:

  1. Plan mode — you can see and edit the plan before any code is written.
  2. Sub-agents / agent teams — you can assign and audit each worker's output.
  3. Process visibility — Perplexity Computer hides the orchestration entirely; Claude Code does not.

The moat matters for the harness decision because the alternative (Perplexity Computer at $176/month) costs more and hides the work. The channel's framing: if you value process visibility (and you should, especially for security-sensitive or money-handling code), Claude Code is the right pick even at a higher per-task cost. The "control freak" framing is the channel's way of saying "we've been bitten by the no-visibility tools before."

Try it yourself

The hands-on goal for this subtopic: pick a Claude Code install path, run one side-by-side against your own non-Claude workflow, and prove to yourself that the harness decision is the one that matters.

  1. Install Claude Code. It's a Node CLI. Default to Sonnet or a cheap model — see Course 2: AI Models for the model-picking framework.
  2. Open a fresh project. Write a one-paragraph brief: a small web app with three features, e.g. "fetch a list of items, render a chart, save a user note."
  3. Send the brief to Claude Code without plan mode first. Note the output: you'll usually get a single sequential implementation.
  4. Reset. Re-run the same brief but with: "Enter plan mode first. Break this into three subtasks and assign each to a sub-agent. Show me the plan before you write any code."
  5. Approve the plan. Watch the sub-agents fan out in parallel.
  6. Time both runs. If the parallel run is faster and the diff is cleaner, you've reproduced the channel's headline result: the gap between Claude Code and Perplexity Computer is mostly configuration.
  7. Compare to Mavis, Hermes, or OpenClaw on the same brief. If you have time, run the same brief through Mavis (§1.1), Hermes (§1.2), or OpenClaw (§1.6). Note where Claude Code wins (low-friction install, fast first output) and where it loses (no 24/7 agent, no Discord bridge, no daily-briefing cron).
  8. Write the one-sentence harness decision. This is the §1.4 capstone. The sentence is: "I'll use [harness] for [task class] because [reason]."

Common pitfalls

  • Skipping plan mode on multi-step builds. This is the single biggest mistake the channel flagged. Default Claude Code is single-track and will brute-force a parallel problem unless you tell it not to. Always force plan mode + agent teams before any non-trivial build.
  • Assuming Pro = Max on Perplexity. Pro subscribers are locked out of Perplexity Computer — only the $176/month Max plan unlocks it. Don't pay for a Pro upgrade expecting the new product.
  • Reading "Opus 1.6" as a typo. In the Perplexity Computer video, the creator refers to Opus 4.6 and "Opus 1.6" interchangeably — they are the same model. Don't treat them as two different releases.
  • Picking Claude Code for a 24/7 agent. Claude Code is a CLI, not a daemon. If you need a 24/7 agent that handles email, file lookups, daily briefings, and Discord chats, Mavis (§1.1), Hermes (§1.2), or OpenClaw (§1.6) is the right harness.
  • Running Claude Code on a Mac Mini tied to your Apple ID. The Mac Mini caveat from nhDA7tcQtx0 applies to every harness that touches a terminal with system access. A $3/month VPS gives you a clean reinstall-OS button that the Mac doesn't.
  • Paying Opus prices for routine maintenance work. Adaptive thinking (from the OpenClaw 3.1 release, 4K1JRI7xA08) is the fix. Route cron jobs and dashboard work to Haiku or Sonnet, save Opus for security-sensitive or money-handling code.
  • Forgetting the long-context gotcha. The 1M-token context window on Opus 4.6 is a separate article in Course 4 §4.4. Default to 200K. Scale to 1M only for full-repo analysis or multi-contract legal review, and only when output quality at the smaller size is actually insufficient.
  • Reading the Perplexity Computer headline as a verdict. The channel framed it as a "KILL" video, then walked the framing back in the follow-up "Claude is back" videos. Don't pin a long-term harness decision on a single headline.
  • Skipping the OpenClaw 3.7 → 3.8 provenance work if you want audit surface. The claude-code track in OpenClaw (covered in Course 4 §4.3) gives every coding task a session ID plus origin tracing. If audit surface matters, route Claude Code through an OpenClaw agent — don't run it standalone.

Sources