Transcript
0:00i work with multiorgastation and this has something to do with my tokenlimit.
0:04in this video I tell you how to save tokens.
0:07one of the most common questions among my videos is
0:10how best to save your tokens,
0:12because the plans are used up so quickly.
0:15so let's start with the 1st of all the most important things.
0:18writes gladly with and you can watch the video again and again.
0:20the 1st point is the theme of orchestration.
0:24orchestration means,
0:25that you are working on one task in different terminals.
0:28that you say the one,
0:30the one terminal is the orchestrator who works with the highest model.
0:34for example until recently Fable
0:37and there now it is opus 5 point 5 with which you can orchestrate super
0:41and the programming itself then made Opus.
0:43so now you can orchestrate but also program with opus.
0:47but if you tomorrow
0:49completely new model comes out Fable 5 or so um Fable 6,
0:53then it would
0:54the orchestrator would work with the highest model
0:56and programming would then make the second-best model.
1:00for example opus5 point 5.
1:01text representation would always be orchestrated with so nice
1:05who so nice can write good texts
1:07and um does not cost that much.
1:10the next item is the topic of sessions,
1:12when uh context.
1:14when you lose sight of the topic of context,
1:16it costs you a lot of money,
1:18if you have already programmed a session very very long
1:21and it is at 80 percent
1:23and you're done with the task.
1:25Make clear and start the next task.
1:28reading the new task costs less than bringing along the 80 percent.
1:33Carrying these 80 percent along costs you
1:35vast amounts of tokens with each input.
1:38therefore ensures,
1:39that then your session is clarified for a new task
1:42or at least compressed with compact Slash compact.
1:47then we come to the point.
1:48is n bit of self-promotion but it is just so
1:50um i have developed my own codecard
1:52and the codecard provides so-called state code analysis
1:56Error does not need to find the LLM
1:58but find 1 tool that costs no money.
2:01Code Card installs you the local um static code analyses
2:05and they analyze Bang Code immediately OB because
2:08there are errors in it and tell the L.
2:10M the L. L.
2:11M can then intervene directly correct the thing and ready.
2:15what is expensive?
2:16the backing rounds are expensive so if you make mistakes.
2:20the software is built
2:21you test through.
2:22oh something doesn't you have to give them the info now
2:25what don't you have to answer again and again,
2:27always on it,
2:27always on it.
2:28so n Diba club can often last an hour and you do not find the error
2:33corrected something makes n new error pure.
2:35these debagging loops cost a lot of money.
2:391 more point is,
2:40you can write in your Cloud Me,
2:43that he should behave short andly um.
2:47This means that there is a clear instruction.
2:49The L. L.
2:50M gives out less information.
2:51is always important to pay attention to,
2:54short and.
2:55statements often,
2:55of course,
2:56lead to the swallowing of information that is necessary for the L.
2:59L. M.
2:59are perhaps important.
3:01this could go to the code quality,
3:02so you have to be careful.
3:041 other trick is,
3:06if you know,
3:07that a very specific file is affected,
3:09i.e.
3:10for example the
3:11Move point J s for the for and you know that there is 1 error in it,
3:16drag the file directly From your From your Visual Studio
3:20into the console and say
3:21here is the error in it.
3:23fix the and the error.
3:24if you make this effort,
3:25he does not have to search the file
3:27and definitely saves tool,
3:29news and tokens.
3:311 other important point is the topic of plane fashion not just build loose.
3:36use Superpowers or Code Guardian here.
3:39have everything planned very very carefully before being implemented.
3:43and very importantly 1 review of your plan you can also do with sonnet.
3:48that means you can create a plan with Opos 5
3:51and then you can make 1 review with sonnet.
3:54I even let Cold Guardian make 1 fruit 5 point 5 review
3:59and 1 Sonnet Review.
4:00that is,
4:01the review is reviewed again,
4:03why invents even then always 12 errors.
4:07and if you already have mistakes in the planning,
4:09think about how expensive it will be later in the reshuffle.
4:13then 1 important point,
4:14which is also extremely important for the topic of debagging and token waste,
4:19make sure that your system architecture,
4:20your software is built this way.
4:22that he has activated.
4:24everything that is done is lured,
4:26logged and Locks Debugs must be enabled.
4:30that is the debug mode
4:31must be activated in your local entwicklungsumgebung .
4:34in diebach mode errors are written directly into the locomotive case
4:38or output and it LLM sees the error directly.
4:41if you have errors do not write there does what not,
4:44but give it exactly the mistake you see.
4:46It's those little things that save a lot of money in the end,
4:50if it LLM does not have to search first,
4:52maybe has a mis and then corrected in the wrong place.
4:56and of course now final one of the levers
4:59is the theme evert when you enter Slash Erfurt.
5:02with Cloud Code the topic of evert level is just important.
5:05many work only with m high or X high or Max
5:09äh for many tasks is also sufficient completely low.
5:12only for programming I definitely recommend X high
5:16I would go to the best possible position.
5:18ultracoat can be done if you want to have many many processes running,
5:23but X high is a good level for me.
5:25Max is something I would use only for the orchestrator,
5:28for example.
5:29with it you can achieve 1 best possible result in any case.
5:33at the orchestrator evaluates yes the inputs and outputs
5:36and that sets priorities.
5:38it does not cost so much tokens
5:39such as programming 5,000 lines of code
5:42who run over Max Everett every time.
5:44here is one of the biggest.
5:46Lever for your Token Limits.
Mind Map
Loading mind map…
Viral Breakdown
Hook (first 3 seconds)
- Verbatim: „i work with multiorgastation and this has something to do with my tokenlimit."
- Hook-Muster: Bold Claim + Insider-Spezifikum („Multiorchestrierung" als Fachbegriff, der sofort eine Nische adressiert)
- Warum es stoppt: Der Satz ist grammatikalisch holprig, aber er trifft exakt den Schmerz der Zielgruppe (Token-Limit). Wer Claude Code / LLMs nutzt, erkennt sofort: „Das betrifft mich." Der Begriff „tokenlimit" ist der Scroll-Stopper, nicht die Orchestrierung.
Emotional Rhythm
- Neugier – „one of the most common questions among my videos" → sozialer Beweis, dass das Problem real ist
- Autorität – „let's start with the 1st of all the most important things" → Struktur-Versprechen
- Spannung/Druck – „Carrying these 80 percent along costs you vast amounts of tokens" → Verlust-Angst
- Twist/Self-Promotion – „is n bit of self-promotion but…" → ehrlicher Bruch, der Vertrauen erzeugt
- Eskalation – „these debagging loops cost a lot of money" → wiederholter Kosten-Frame
- Climax – „here is one of the biggest. Lever for your Token Limits." (Effort-Level-Sektion am Ende)
- Erleichterung – konkrete Handlungsanweisungen (Slash compact, Datei reinziehen, Debug-Mode)
Climax-Moment: Der Effort-Level-Abschnitt („X high is a good level for me… Max only for the orchestrator") – hier wird das größte Sparpotenzial offengelegt.
Keyword Density
- tokens / Tokenlimit – algorithmischer Reach-Treiber (Suchbegriff der Nische)
- orchestration / orchestrator – Fach-Identität, zieht Power-User
- save / sparen / costs – emotionaler Pull (Geld-/Verlust-Angst)
- session / context – Problem-Keyword, hohe Suchintention
- debugging loops / debagging – Schmerz-Keyword, triggert Wiedererkennung
- plan / review – Lösungs-Keyword
- effort / X high / Max – konkreter Handlungs-Trigger
- Claude Code / Opus / Sonnet – Tool-Keywords, hohe Suchrelevanz
Algorithmus vs. Emotion: „tokens", „Claude Code", „Opus" treiben die Auffindbarkeit. „save", „costs", „money", „expensive" treiben die Verweildauer.
Why It Spreads
- Pain-first Positioning: „the plans are used up so quickly" – exakt der Satz, den die Zielgruppe denkt, bevor sie das Video sieht.
- Listenformat mit nummerierten Levers: „the 1st point…", „the next item…", „1 other trick…" – jeder Punkt ist ein eigener Screenshot-/Rewatch-Moment.
- Konkrete, kopierbare Befehle: „Slash compact", „drag the file directly… into the console", „Slash Erfurt" – Zuschauer speichern das Video, um es später auszuführen (High Save-Rate = Push).
- Anti-Hype-Ehrlichkeit: „is n bit of self-promotion but it is just so" – bricht die Werbe-Skepsis, erhöht Kommentar- und Share-Bereitschaft.
- Verlust-Framing als Wiederholungsanker: „costs you vast amounts of tokens", „cost a lot of money", „expensive" – dieselbe Emotion in jedem Abschnitt, hält Retention hoch.
What You Can Steal
- Eröffne mit dem Schmerz-Keyword deiner Nische, nicht mit deinem Namen. „tokenlimit" statt „Hi, ich bin…".
- Baue das Video als nummerierte Lever-Liste, bei der jeder Punkt einzeln screenshotbar ist – das erzeugt Saves und Rewatches.
- Platziere den stärksten Hebel ans Ende (hier: Effort-Level) und kündige ihn mit „one of the biggest levers" an – das hält Zuschauer bis zum Schluss.
