DON'T WASTE SO MANY TOKENS! PI CODING AGENT vs OPENCODE with same local LLM.
Luigi Tech runs a head-to-head between the Pi coding agent and OpenCode using the same local model, Qwen 3.6 35B-A4B, on the same bug-fix task in a vibe-coded 3D Tic Tac Toe game, timing both and comparing context usage. Both harnesses produce the same working fix, but Pi finishes in 7 minutes 44 seconds with roughly half the tokens of OpenCode's approximately 12-minute, 23K-context run, leading to the conclusion that the model matters more than the harness.
Luigi Tech15 minTranscript found
Quick learning frame
Read this before watching.
Creative automation uses agents to accelerate production while keeping human taste in story, pacing, selection, and critique.
New playlist item from Luigi Tech; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Skill you build: The ability to benchmark coding-agent harnesses fairly, same model, same repo state, same prompt, with a timer and token counts, and to judge where harness overhead is worth it versus wasted context.
Watch for the shift from claim to mechanism. The learning value is the point where the transcript reveals a repeatable action, tool boundary, context move, review habit, or artifact.
Concept diagram
Where this video fits.
01Brief
02Source
03Generation
04Selection
05Edit
06Taste Review
Deep lesson
Turn this video into working knowledge.
1,204 cleaned transcript words reviewed across 467 timed caption segments.
Thesis
DON'T WASTE SO MANY TOKENS! PI CODING AGENT vs OPENCODE with same local LLM. teaches a practical creative automation move: Luigi Tech runs a head-to-head between the Pi coding agent and OpenCode using the same local model, Qwen 3.6 35B-A4B, on the same bug-fix task in a vibe-coded 3D Tic Tac Toe game, timing both and comparing context usage. Both harnesses produce the same working fix, but Pi finishes in 7 minutes 44 seconds with roughly half the tokens of OpenCode's approximately 12-minute, 23K-context run, leading to the conclusion that the model matters more than the harness.
The goal is not to remember the video. The goal is to extract the operating principle, tie it to timestamped evidence, test how far the claim transfers, and make something reusable.
1:31
Controlled harness test
“files. So I have index HTML, game.js and we will try the same prompt inside both harnesses and we will compare the results and I also will use a timer to see how much it will take to...”
The experiment fixes two real bugs in the game, cell cubes packed too close together and winning markers not highlighting in green, using an identical prompt in both harnesses, the repo reset between runs, and a timer running. The model is Qwen 3.6 35B-A4B, which Luigi calls the best model you can run on your own computer right now, and Pi starts by analyzing the directory of separated source files (index.html, game.js). Design your own two-harness test: pick one small real bug, write a single shared prompt, and record time, context used, and whether the fix works for each harness with the repo reset in between.
6:13
Overhead has a cost
“also the winner markers highlighted. So it works. And this was with Pi coding agent. So now we will do the same test with open code and same model and same code. So I will reset the code.”
For the OpenCode run Luigi uses Basico, a minimal custom agent defined in a simple markdown file, to recreate similar conditions, yet OpenCode already sits at 12K context before work starts and adds a to-do list step that Pi skipped. He also notes Gemma 4 26B could recreate the game but failed at tool calling to edit the files, which is why Qwen 3.6 was the only model tested. Inspect your harness's starting context before your first prompt and list which built-in features (to-dos, guardrails, prompt tweaks) are earning their token cost on small tasks.
10:10
Model beats harness
“compared to Pi Agent. But, yeah, Pi Agent was able to do this kind of task also without to do in the middle, but maybe in more um intricated situations, it could be useful to have a to...”
Both harnesses ship the same working result, spaced cubes and green winning markers, but OpenCode takes about 12 minutes and roughly 23K of context while Pi delivered in 7:44 with about half the tokens. Luigi's conclusion: the LLM and the quality of data placed into context matter most; the harness is useful but secondary, and Pi's lower overhead won this use case without a big system prompt. Write a one-paragraph rule for when you would choose a minimal harness over a feature-rich one, using time, token cost, and task complexity as the deciding factors.
01
Brief
Start with this video's job: Luigi Tech runs a head-to-head between the Pi coding agent and OpenCode using the same local model, Qwen 3.6 35B-A4B, on the same bug-fix task in a vibe-coded 3D Tic Tac Toe game, timing both and comparing context usage. Both harnesses produce the same working fix, but Pi finishes in 7 minutes 44 seconds with roughly half the tokens of OpenCode's approximately 12-minute, 23K-context run, leading to the conclusion that the model matters more than the harness. Treat "Brief" as the outcome you are trying to make visible, not a topic label. Anchor it to 1:31, where the video says: “files. So I have index HTML, game.js and we will try the same prompt inside both harnesses and we will compare the results and I also will use a timer to see how much it will take to...”
02
Source
Use "Source" to locate the part of the creative automation workflow the video is demonstrating. Ask what changes in your real setup if this claim is true. Anchor it to 6:13, where the video says: “also the winner markers highlighted. So it works. And this was with Pi coding agent. So now we will do the same test with open code and same model and same code. So I will reset the code.”
03
Generation
Turn "Generation" into the reusable artifact for this lesson: A creative workflow board with critique criteria and review checkpoints. This is where watching becomes something you can inspect and reuse.
04
Selection
Use "Selection" as the application surface. Decide whether the idea touches a browser flow, a local file, a model choice, a source document, a UI, or a review step.
05
Edit
Use "Edit" to prove the lesson. The evidence should connect back to the video title, transcript anchors, and a concrete output, not a generic best-practice claim.
06
Taste Review
Use "Taste Review" to carry the idea forward: save the prompt, checklist, diagram, or operating rule that would make the next agent run better.
Example
Source-backed work packet
Convert the video into a scoped task that includes the transcript claim, target workflow, acceptance criteria, and proof. The output should be a creative workflow board with critique criteria and review checkpoints..
Example
Claim vs. demo brief
Separate what the speaker claims, what the demo actually proves, and what still needs outside verification before you adopt the workflow.
Example
Teach-back module
Transform the lesson into a definition, a mechanism diagram, one misconception, one practice exercise, and a check-for-understanding question.
Do not learn it wrong
Treating the title as the lesson without checking what the transcript actually says.
Letting the prompt drift into generic advice that could apply to any video in the playlist.
Copying the tool setup without identifying the operating principle that transfers to your own stack.
Skipping the artifact, which means the learning never becomes operational or inspectable.
Do not count this as learned until these are true.
01
State the transcript-backed claim in your own words: Luigi Tech runs a head-to-head between the Pi coding agent and OpenCode using the same local model, Qwen 3.6 35B-A4B, on the same bug-fix task in a vibe-coded 3D Tic Tac Toe game, timing both and comparing context usage. Both harnesses produce the same working fix, but Pi finishes in 7 minutes 44 seconds with roughly half the tokens of OpenCode's approximately 12-minute, 23K-context run, leading to the conclusion that the model matters more than the harness.
02
Explain the practical stakes without hype: New playlist item from Luigi Tech; queued for transcript-backed review, topic mapping, and a practical learning artifact.
03
Map the idea onto the Brief -> Source -> Generation -> Selection -> Edit -> Taste Review sequence and name the weakest link.
04
Produce the artifact and include the evidence that proves it: A creative workflow board with critique criteria and review checkpoints.
Put it into practice
Give this grounded prompt to Codex or Claude after watching.
You are helping me turn one specific YouTube video into real, durable learning.
Source video:
- Title: DON'T WASTE SO MANY TOKENS! PI CODING AGENT vs OPENCODE with same local LLM.
- URL: https://www.youtube.com/watch?v=dRialfHHwRw
- Topic: Creative Automation
- My current learning frame: Reproduce the benchmark on your own machine: take a small buggy project, run the identical fix prompt through two agent harnesses backed by the same local model, and compare wall-clock time, context consumed, and result quality to decide which harness earns its overhead.
- Why this matters: New playlist item from Luigi Tech; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Transcript anchors from this exact video:
- 1:31 / Evidence 1: "files. So I have index HTML, game.js and we will try the same prompt inside both harnesses and we will compare the results and I also will use a timer to see how much it will take to..."
- 3:13 / Evidence 2: "And but maybe it's more interesting to see the time spent to to fix the game. And yeah, it is working and then we will do the same task with open code and I will reset the repo..."
- 6:13 / Evidence 3: "also the winner markers highlighted. So it works. And this was with Pi coding agent. So now we will do the same test with open code and same model and same code. So I will reset the code."
- 8:14 / Evidence 4: "engine tool which we won't use it in this use case. So it's a very simple agent just to to see just to recreate similar conditions for open code. And we already are using 12k of the context."
- 10:10 / Evidence 5: "compared to Pi Agent. But, yeah, Pi Agent was able to do this kind of task also without to do in the middle, but maybe in more um intricated situations, it could be useful to have a to..."
- 11:56 / Evidence 6: "that then it will finish. And we are at 12 minutes, so it's more. But, okay, it's finished, I suppose. And as you can see, the context used it's uh 23K about with open code and uh probably..."
- 14:48 / Evidence 7: "overhead and we got a good result also without a very big prompt in the LLM. Let me know in the comments which is your preferred open source Arnes coding agent and see you in another video. Bye."
Your task:
1. Use the transcript anchors above as the primary source packet. If you add outside context, label it clearly as outside context and keep it secondary.
2. Create a source-check table with columns: timestamp, claim, what the demo proves, confidence, and what still needs verification.
3. Extract the actual teachable claims from the video. Do not invent claims that are not supported by the title, lesson frame, or transcript anchors.
4. Build a reusable learning artifact: A creative workflow board with critique criteria and review checkpoints.
5. Include:
- a plain-English definition of the core idea
- a diagram or structured model using this sequence: Brief -> Source -> Generation -> Selection -> Edit -> Taste Review
- 3 concrete examples that apply the video idea to real agentic work
- 2 failure modes the video helps prevent
- a checklist I can use the next time I run Codex or Claude
- one practical exercise with a clear done signal
6. Add a "learning transfer" section: what changes in my workflow tomorrow if I actually learned this?
7. Add a "source check" section that cites which transcript anchor supports each major takeaway.
Quality bar:
- Make this specific to "DON'T WASTE SO MANY TOKENS! PI CODING AGENT vs OPENCODE with same local LLM.", not a generic Creative Automation essay.
- Prefer operational examples, failure modes, and reusable artifacts over broad definitions.
- Call out uncertainty instead of smoothing over weak evidence.
- If evidence is weak, say what transcript segment or timestamp needs review instead of guessing.
- Finish with a concise artifact I could paste into my learning app.
Misconceptions
What to stop believing.
Creative AI removes the need for taste.
It increases the need for taste because output volume explodes.
The best prompt is enough.
References, critique, iteration, and post-production matter just as much.
Practice studio
Learning only counts when you make something.
01
Transcript evidence map
Separate what the video actually says from what you already believe about the topic.
3 source-backed takeaways with timestamps, confidence, and a transfer note.02
One useful artifact
Apply the video to a real workflow and produce a creative workflow board with critique criteria and review checkpoints..
A reusable artifact with a done signal and one verification step.03
Teach-back card
Explain the lesson to someone who has not watched the video yet.
A 90-second explanation, one diagram, one example, and one misconception to avoid.
Recall check
Answer first, then reveal — without rewatching.
What two bugs was each harness asked to fix, and with which local model?
Why was the comparison limited to Qwen 3.6 instead of also using Gemma 4 26B?
What was the final verdict on harness versus model after both runs finished?
Source shelf
Use the video as a doorway, then verify with primary sources.
