Com organitzar la feina amb ChatGPT Work i GPT‑5.6: correu, calendari i Sites
Un sistema pràctic per connectar aplicacions, preparar reunions, programar revisions i publicar Sites, amb els límits de permisos, context i supervisió que cal tenir presents.
Peter Yang mostra com ha convertit ChatGPT Work i Codex en una mena de sistema operatiu personal: connecta correu i calendari, crea instruccions de veu, prepara entrevistes, programa revisions setmanals i publica guies o prototips com a Sites. El vídeo és una demostració potent, però algunes recomanacions són preferències personals i la distinció entre productes és més profunda que un canvi d’interfície.
Chat, Work i Codex no són exactament el mateix
Al minut 1:04, Yang defineix Chat com la conversa habitual, Work com ChatGPT dins d’altres aplicacions i Codex com una versió amb eines per a desenvolupadors. Després afirma que Work i Codex només es diferencien gairebé per la pàgina de pull requests.
La documentació oficial els separa per tipus de feina:
- Chat és per a preguntes, cerca, ideació i ajuda ràpida.
- Work és un agent per a recerca i lliurables llargs: documents, fulls de càlcul, presentacions, informes i Sites.
- Codex està dedicat al desenvolupament: repositoris, fitxers locals, terminal, proves, revisió i canvis de codi.
La història també és diferent. Les converses cloud de Work se sincronitzen entre web, mòbil i escriptori. Codex manté una vista i un historial separats; al mòbil s’hi accedeix a través de Remote quan la sessió és compatible. Triar una superfície no és només qüestió de gust: determina quin entorn, quins fitxers i quines eines pot tocar l’agent.
Quin GPT‑5.6 convé utilitzar
Al minut 2:22, el vídeo recomana GPT‑5.6 Sol amb esforç mitjà com a punt d’equilibri i reserva l’esforç alt per planificar. És una heurística raonable per al seu ús, no una regla universal.
OpenAI descriu Sol com el nivell de més capacitat, Terra com l’opció equilibrada i Luna com la més ràpida i econòmica. En Work i Codex, la disponibilitat depèn del pla i del desplegament al compte. L’elecció correcta depèn del coll d’ampolla:
- Luna per a transformacions simples i iteració ràpida;
- Terra per a treball quotidià amb bon equilibri;
- Sol per a ambigüitat, recerca o decisions complexes;
- més esforç quan una resposta incorrecta costa més que uns minuts o crèdits addicionals.
Per saber-ho de debò, cal provar la mateixa tasca amb criteris de qualitat, temps i consum. La sensació que dos nivells «fan el mateix» en un exemple no mesura què passa en un conjunt variat.
Instruccions personals: preferències, no una autobiografia infinita
Al minut 3:18, Yang configura instruccions perquè l’agent expliqui conceptes tècnics sense argot, sigui franc, escrigui en veu activa i eviti tics de text generat. És un bon ús: són preferències duradores que s’apliquen a moltes tasques.
Les instruccions funcionen millor quan són breus, no es contradiuen i es poden observar. «No facis servir blocs de codi per a prosa» és verificable; «escriu perfecte» no ho és. Informació específica d’un projecte ha d’anar al projecte o al repositori, no a una configuració global.
Tampoc s’hi haurien d’enganxar secrets, claus o dades personals innecessàries. La personalització ajuda a mantenir el to, però no substitueix revisar un correu important ni comprovar fonts.
Connectar aplicacions amb el mínim permís útil
Al minut 6:23, el creador recomana instal·lar Gmail, Calendar, Documents, Presentacions, Fulls i Drive, i afirma que no veu gaire inconvenient a tenir molts connectors.
Aquí convé ser més prudent. Cada plugin amplia el context i la superfície d’acció: pot llegir dades, proposar canvis o executar-los segons els permisos i les aprovacions. La configuració responsable és:
- connectar només el servei necessari per al flux actual;
- revisar compte, domini i permisos abans d’autoritzar;
- exigir confirmació per a enviaments, esborrats, compres o canvis públics;
- desconnectar integracions que ja no s’utilitzen;
- separar comptes personals i corporatius quan calgui.
Organitzar un projecte per tema i un fil per flux recurrent pot conservar context, tal com fa Yang. Però un fil etern també acumula supòsits antics. Quan canvien requisits o permisos, és millor iniciar-ne un de nou amb una síntesi explícita.
Un «cap de gabinet» per al correu
Al minut 9:30, Yang demana revisar els últims tres dies de correu, extreure cinc accions pendents i proposar llistes de què donar-se de baixa. Després fa que el navegador obri missatges i executi les baixes.
El patró és útil perquè separa descoberta i acció:
- primer, resumir i enllaçar les evidències;
- després, decidir què cal fer;
- finalment, executar només amb aprovació.
Per a correus, l’agent pot confondre un butlletí amb un missatge contractual o interpretar malament el silenci d’un interlocutor. Abans d’enviar, cal revisar destinatari, fil, adjunts, dates i to.
Yang també fa analitzar missatges enviats per crear una skill que redacti amb la seva veu. La prova amb un correu real és la part important: una descripció d’estil no garanteix que la resposta sigui adequada per a clients, amistats i proveïdors. Convé incloure exemples per context i prohibir que la skill inventi compromisos.
Calendari i preparació de reunions
Al minut 13:47, el vídeo programa una videotrucada, detecta un conflicte i prepara entrevistes de podcast. El flux combina calendari, recerca, una skill pròpia i un document compartit.
La idea general és bona: usar l’esdeveniment com a desencadenant per reunir biografia, articles, preguntes i logística. Però reservar una hora és una acció externa. Zona horària, durada, convidats, privacitat, recurrència i enllaç de reunió s’han de confirmar abans de crear o modificar l’esdeveniment.
També és útil separar un dossier intern d’una guia que rebrà el convidat. Notes crítiques, dades personals o hipòtesis de recerca no haurien d’acabar automàticament en el document compartit.
Automatitzar només després d’haver provat el flux
Al minut 17:19, Yang programa una tasca cada divendres: revisar correu, mirar set dies de calendari i preparar documents per a episodis de podcast.
La seqüència és encertada: primer executar manualment, corregir resultats i després programar. Una automatització multiplica tant els encerts com els errors. Ha de tenir:
- abast temporal i fonts definides;
- sortida clara i lloc on trobar-la;
- límits sobre accions irreversibles;
- notificació només quan hi ha novetats;
- procediment per pausar-la o corregir-la.
La documentació oficial adverteix que una tasca programada creada dins d’un projecte no pot accedir als fitxers d’aquell projecte. Per tant, cal provar el context real que tindrà quan s’executi sola i no assumir que veurà el mateix que el fil interactiu.
Continuar des del mòbil
Al minut 19:23, el creador obre Remote al mòbil i manté el Mac despert perquè les sessions locals continuïn disponibles. Aquest requisit s’aplica al seu flux de Codex connectat a l’ordinador. Les converses cloud de Work, en canvi, se sincronitzen directament.
Deixar un portàtil actiu amplia el temps d’exposició. Cal bloqueig de pantalla, xifratge, alimentació segura i permisos limitats. I el mateix Yang assenyala un risc menys tècnic: supervisar agents constantment pot convertir-se en una nova forma de distracció.
Sites: de document a experiència interactiva
Al minut 21:01, el vídeo mostra una guia de viatge al Japó, un joc de multiplicacions i un PRD interactiu amb requisits, disseny i esquema de dades. Sites fa més fàcil compartir un resultat com a web que com una llista estàtica.
La interactivitat és especialment útil per a prototips, taulers i documents de producte. No elimina la revisió abans de publicar. Cal comprovar dades personals, llicències d’imatges, accessibilitat, enllaços, funcionament mòbil i si l’URL serà públic. Un prototip generat en minuts tampoc equival a un producte amb autenticació, seguretat, còpies de seguretat i manteniment.
En resum
El sistema de Yang funciona perquè combina context persistent, connectors, skills provades i automatitzacions que parteixen d’un flux manual. La seva millor idea és demanar un resultat concret i revisar-lo abans de delegar-lo de manera recurrent.
Els matisos són igual d’importants: Work i Codex tenen àmbits diferents, el model «ideal» depèn de la tasca, més plugins impliquen més permisos i un Site ràpid no és necessàriament una aplicació preparada per a producció. Un bon sistema personal no és el que automatitza més, sinó el que deixa clar què pot llegir, què pot canviar i quan ha de demanar permís.
Contrast i context
Fonts consultades
- 01
-
02
OpenAI ChatGPT Work and Codex
-
03
OpenAI GPT-5.6 in ChatGPT
-
04
OpenAI ChatGPT Work for every team
Font de treball
Transcripció amb marques de temps
Consulta la transcripció
-
0:00
, obre el vídeo en una pestanya nova
Hey everyone, I now use Chat GPT Work and Codex for almost everything that I do on my computer and I'm going to show you exactly how to do the same in this video. Ever since OpenAI launched GPT-5.6, the Chat GPT desktop app has been growing like a weed. It literally is growing by a million users every single day. But let's answer some burning questions first, right? Number one, when should you use Chat, Work, or Codex? Which GPT-5.6 model is right for you? How do you connect Codex to all the favorite apps that you actually use? And how do you keep projects and threads organized as you have more chat conversations? How do you build an AI chief of staff in Codex to manage your emails and calendar? And how do you use Codex on your phone? And finally, how do you build beautiful websites for free and get Codex to work for you while you sleep? I'm going to cover all of these topics step-by-step in this tutorial. It's going to be super beginner-friendly, so let's get right into it. First, what exactly is the difference between Chat, Work, and Codex? Well, basically it boils down to this. Chat is the regular Chat GPT that we're all familiar with. Work is Chat GPT, except it can now work inside your favorite apps. And Codex adds features designed for coding and developers. Honestly, I'm not a huge fan of this Chat GPT Work and Codex distinction. As far as I can tell, the main difference is the UI and how responses are presented. So, let's take a look. So, here we have the Chat GPT desktop app. And you can see up here that we can switch between Work and Codex, right? And pretty much the only difference, if you look at this, is when I switch to Codex, there is now a pull requests page here. And pull requests, by the way, are basically what developers use to review and approve each other's code. And if I switch back to Chat GPT Work, that part is missing. But as far as I can tell, that's pretty much the main difference between ChatGPT work and Codex. In my opinion, these two things should exist as one thing because Codex itself is not too difficult to understand and use. It's basically just a bunch of chat threads except now you can actually do stuff. Personally, I prefer to stick to Codex, but ChatGPT work works perfectly fine, too. Okay, now let's talk about something that's even more confusing. So, GPT 5.6 comes with three different models: Soul, Terra, and Luna. And each option has five different effort levels from low to ultra. So, what model should you use? Well, here's a chart that shows intelligence versus output tokens for GPT 5.6. We're looking for higher intelligence but fewer output tokens. And you can see here that the sweet spot is GPT 5.6 Soul medium effort. It delivers good enough intelligence without burning all your tokens. And sometimes I use Soul with high effort for planning here.
-
3:00
, obre el vídeo en una pestanya nova
But overall, I really haven't noticed much of a difference between the two. And I almost never use Terra or Luna. So, in my opinion, you should just stick with GPT 5.6 Soul with medium effort to get good enough intelligence while still conserving your tokens. Now, let's jump back into the ChatGPT desktop app. If you're using ChatGPT on the web, the desktop app is the biggest AI native upgrade you can do. You can just do so much more on the desktop app than on the web. And the way to install it, just search for ChatGPT app install on Google and you'll find it. So, the first thing I'd like to do on a fresh install is to go to settings, personalization, and change the custom instructions. So, let's go to settings. Let's go to personalization. You can see here that we have the custom instructions here. Let me expand it over here a little bit more. The reason I like to do this is because I like to customize how the model responds to me. And there's things that I want the model to understand. So, let's kind of walk through this. The first bullet is telling the model about me, right? I have over a decade of product experience, and I'm also a creator. I want to learn technical concepts. That's something I'm working on. So, while I'm building with Codex and ChatGPT work, I like to have the model actually teach me system design, technical trade-offs, how APIs work, and so on and so forth. And I like to tell it to avoid jargon and explain everything in easy-to-understand way. So, that's number one. Number two is I have a pet peeve where I don't like the model to use code blocks for anything except for actual code. It tends to want to use code blocks for just like regular prose, which I really don't like. So, that's there. And number three and a four, I think is pretty important. So, number three is be candid and thoughtful. Tell me what I need to hear while staying curious and caring. And one thing that I've noticed, even with the best models like GPT 5.6, is that it tends to just want to agree with you. It's kind of like rewarded by just always agreeing with what you what you think. And what I want the model to do is actually to stick to principles and push back when it needs to, right? Okay, so number four is write plainly in the active voice, minimize hedging, padding, and grand claims, and avoid all these AI slop terms. I really don't like "This isn't about X, this is about Y." And a lot of em dashes, a lot of just like throat clearing stuff. Especially since I use ChatGPT a lot to edit my writing, so having this kind of stuff here is very important. And by the way, if you check out my GitHub, which I've linked in the description, there is a free no AI slop tool that goes into way more than these things. It actually removes over 20 pieces of AI slop phrases to help you edit any piece of writing. So, check that out in the video description. And then there's some stuff down here, and I don't think you actually need this
-
5:50
, obre el vídeo en una pestanya nova
stuff. The reason I have this stuff here is because I used to use Claude code a lot, and I still do sometimes, and I want to make sure that I have one canonical set of skills between Claude code and ChatGPT work and Codex. So, there's something called SimLink to make sure that both apps refer to the same folder. But, if you just use Codex, you don't actually need five and six. The point here is that you want to add some of these custom instructions to personalize how the model responds to you. The next thing I'd like to do, again, on a fresh install of the ChatGPT app, is go to plugins. Now, plugins are extremely important to let ChatGPT actually work inside your favorite apps. And I think the bare minimum set that you should install is Gmail, Google Calendar, Documents, Presentations, Spreadsheets, and Google Drive, right? So, that ChatGPT can actually manage your emails, calendar, and edit your docs, presentations, and sheets. Let's take a look at all the plugins that I have installed. So, here they are. I have over 23 plugins installed. And then, other than what I mentioned, there's the PDF. You know, I really dislike using Adobe Acrobat and any Adobe product, to be honest. So, being able to let ChatGPT read and also edit my PDFs, it's just incredibly useful. And scrolling down here, there's other stuff around Slack for community messages, Notion for my wiki, Figma for design, and RemNote for my meeting notes, right? So, as far as I can tell, there isn't any real drawback to installing a ton of plugins. So, definitely install plugins for all the apps that you work with on a daily basis. This is the key setup step that makes Codex incredibly powerful. So, I cannot reinforce this point enough. Okay, now that we covered plugins, let's cover projects and threads. So, I have two pinned threads here, and if I expand this, here are all my projects, right? And as you can see, I have projects for my personal operating system and various products and apps that I'm working on down here. Now, within each project, I like to organize threads based on individual tasks and workflows. So, for example, in my personal OS project, I have threads for making podcast episodes for various guests, for helping to edit my newsletter post, drafting and posting social posts across all social networks, and so on and so forth, right? And one thing that Codex does really well is it's really good at compacting previous messages in the thread while still remembering what they were about. So, basically, that allows me to essentially just have extremely long-running threads. You can see this is a very long thread with a lot of back-and-forth conversation, and I just keep going. I'm not worried about Codex getting confused or not knowing what's going on. Another benefit of of having one thread per workflow is that it's just so much more easier to
-
8:48
, obre el vídeo en una pestanya nova
manage than having to manage a whole bunch of threads for individual tasks, right? Now, to organize your threads better, you can also drag and drop individual threads inside a project, and of course, you can pin threads to the top that you're working on. One more thing is if you hold down the pound sign on your Mac, you can quickly navigate between different threads. So, I can just go to pound three, pound four, pound two, and so on to now navigate. So, that's basically threads. The TLDR here is that I recommend all of you use long-running threads, one thread per workflow. And now that everything's organized, let me show you how to actually get some real work done. Okay, so all of us have to manage emails and calendar meetings, right? And these days, I don't actually go to Gmail and Google Calendar much. Instead, I try to manage everything through Coda. So, let me show you exactly how that works. So, first of all, you need to make sure that you have the plugins installed for Gmail and Google Calendar for this to work. And then, what I recommend you do is you start a chief of staff thread. And the way you start a thread is go to new task and just say, "Let's make this the chief of staff thread." Okay. So, now that we have this thread in place, one thing that I frequently miss with my emails is I read the email and I don't follow up right away and I just forget to follow up, right? So, what I did is I started by asking Coda, "Check my email from the past 3 days and list the top five open action items to follow up on and emails to unsubscribe from. And also link the actual emails. Do not share any confidential information because we're giving this tutorial." So, here's the list that Coda came up with. It looks like there is a new device logging, there is a Google payment issue, there is an automation that I built that broke, there is some thread about granola and a meeting with Brex, right? And here is a bunch of unsubscribe candidates. And if I click one of these things, it'll take me to the actual email. So, what's the next step here? Now that we have this list, what we can do is basically just tell Coda, "Hey, why don't you unsubscribe from all five emails?" And because Coda can use the browser, it can actually open each of these individual emails in Chrome and click the unsubscribe button and just get it done. So, it looks like it worked for 2 minutes and it unsubscribed to all five emails that I wanted to unsubscribe from, which is pretty incredible, right? Because previously, I had to go into the emails and click unsubscribe one by one and there's always new marketing emails coming in. I also use Coda all the time to find and follow up on previous emails. So, for example, I sent an email to book a hotel in Japan and they haven't replied to me yet. So, I basically ask Codex to find the email for the Yufuin Hotel in Japan to send a follow message,
-
11:42
, obre el vídeo en una pestanya nova
but let me review it first. So, found the email thread and it said that I sent the last follow message a week ago and it went off and actually drafted a reply. So, I'm following up on my reservation. We want to stay during Christmas time, prefer this room, and so on so forth. And because and because we have our Gmail extension hooked up to Codex, I can just tell it to send the email after I edit and review it. So, I just said send and it got sent successfully to the existing thread. Now, to take this to the next level, what you can do is you can train Codex to reply to emails using your voice and tone. If you look at this email, it's a little bit too formal. So, why don't we actually get Codex to learn my voice and tone for emails? So, basically I told it to review my sent messages and let's create a slash email skill to draft emails in my voice and style. And then it worked for 4 minutes. Let's take a look at what actually did. It went through a bunch of emails and reviewed everything. It actually even recognized that my voice is different whether I know the person versus someone new, right? And then there there's something here. And it basically created a skill here. Let's take a look. So, created a skill called email and here's the workflow and here is the voice that it should follow. Now, let's actually test the skill with the Japan email draft that we had. And 30 seconds later, you can see here that it drafted the email that's a little bit shorter and less formal, which is more casual like my style. So, this is important, guys. You notice how I asked Codex to do some research on my writing style for emails. It put together a skill for me, right? And then I actually tested the skill to make sure that it's matching my voice before I just kind of said, "Okay, let's keep using this from now on." Like you always want to test the skills that Codex makes to just make sure that it actually is what you want, right? Cuz usually it doesn't really get it perfect in one shot. You got to iterate and go back and forth with it and make it better. And I have another video about building self-improving skills that I'll link in the video description. Now that we talked about email, let's quickly talk about using Codex to manage your calendar as well. You know, I always found it a pain to open Google Calendar and manually schedule an event by clicking a bunch of buttons and entering a bunch of text. It's much easier to just ask Codex to book calendar meetings for you. So for example, I asked it to book a Google Meet meeting between me and Char for Friday at 9:00 a.m. PST called Peter Char for 15 minutes. And Codex basically went off and it found Char's email, right? And it even found that I had a conflicting meeting where I had to drop off my kid at her camp during the same time. And eventually booked the calendar event here and even added a Google Meet link. If you're booking a meeting to for example meet someone for coffee or lunch, you can just mention the name of
-
14:38
, obre el vídeo en una pestanya nova
the restaurant and Codex will find it for you. Or if you're having like just kind of like a WhatsApp conversation with someone and you agree to a meeting time and place, usually what I do is I just screenshot the conversation and copy it directly into Codex and it's able to book the meeting for me. So these are just like super convenient, especially if you do it from your phone. Another thing that I did recently a few weeks ago was I just asked Codex to add all upcoming World Cup matches starting from the quarterfinals onwards and book it on my calendar as optional events. And it did a bunch of research and basically added all the matches for me. For me to do this manually, it would have taken me at least 5 10 minutes to do, right? But Codex basically did it in 1 or 2 minutes. Now, let me show you guys something more advanced they can do on your calendar. So, what I asked Codex now is find and list all upcoming podcast meetings in the next 7 days cuz I do a lot of podcast interviews. So, it basically found that I had three podcast interviews. There's one today with Jason from Codex, Hamel, and then Paul. Now, I have a podcast prep skill that I use to prepare for each episode by doing research online, drafting a few questions to ask the guest, and then putting together an interview guide. So, basically I just ran podcast prep for Paul, which I'm interviewing on Thursday. And it's asking me a bunch of questions about what do you most want viewers to learn from Paul? And I have some reference material and Paul's articles on X that I pasted here. And then it went off and did 4 minutes of work. So, it did a bunch of research, it searched a bunch of stuff, and it came up with a research-backed outline, right? Like you see here. And with these interviews, I found it best to figure out the packaging, like basically the YouTube title and thumbnail first, so you can figure out what the right angle is before you actually do the interview. And here we have a bunch of different options. And number one is Codex runs my app business because Paul has built a lot of iOS and Mac apps using Codex. So, that's the main topic, right? And based on that, it went off and wrote a complete guest-facing draft guide right here. So, you can see here that there's a welcome, there's questions that I should ask, there's sample answers, and so on and so forth. Normally, I would go back and forth with this a couple of times to make sure that it's actually good and apply my taste and style to it. But, in this case, I just asked it to go ahead and add it to this Google document. So, let's open a Google document now. As you can see here, that now I have a full interview guide for Paul that I can actually share with him, which is incredibly useful. So, where am I going with all this? Basically, pretty much every single week I have one or two or three podcast interviews that I have to prepare for. And this is how you get Codex to do work
-
17:28
, obre el vídeo en una pestanya nova
for you proactively, right? So, down here I have a prompt saying "Schedule a task to run every Friday at 7:00 a.m. And this time we're going to merge all the email and calendar workflows that I just showed. So, number one, check my past three days of email and list the top five open action items and the emails to unsubscribe from. And number two, check the next seven days of calendar events. And if you find a podcast episode, it should say "Podcast between Peter and X", then run the skill and create and link the docs for each episode. And also check for other important meetings that I should prep for." So, basically, created this scheduled task. And let's open it here. I can see here there's a bunch of description and so on and so forth that Codex came up with. And if we go here to the schedule page, you can see that there is an active Friday chief of staff brief that we just scheduled. And I have some other things here to clean up my local downloads folder and get some funnel data and so on and so forth. So, let's go back to our chief of staff thread. And I just want to emphasize, people who are really good at using Codex, they have a ton of scheduled tasks going. Right now, we have scheduled a weekly task, right? I've talked to folks on the Codex team, they have pretty much daily scheduled tasks going that look at their Slack, their emails, Granola transcripts, and more, and trying to help them prioritize and figure out what they should be doing. So, this saves them a lot of time from having to check five different apps to get the enough context to figure out what they should do next. So, definitely make use of scheduled tasks. It's incredibly valuable. So, basically, the lesson here is after you do the manual work of Codex, right? We did the manual work to check your emails, find action items, check your calendar events, we gave it some feedback. After you do this manual work, usually what you want to do is ask Codex to schedule a weekly task, a daily task, or even a monthly task so that it can do this work more proactively in the future. Okay, now let's talk about using Codex from your phone because we often have to manage our emails and calendar there, and we also are often on the go. So, to get access to Codex from your phone, just open the ChatGPT app like I have here and click this thing and then go to remote. So, remote is basically all your Codex threads, and it basically talks to Codex through your computer. So, there's a really big catch here, which is your computer has to be on for all these remote threads to actually work. So, how do you keep your computer on if you're not around? Well, the way I do it is I basically have this app installed on my Mac called Amphetamine. And it's a basically a keep-awake utility, right? So, if I open Amphetamine here, I'm able to set how many hours and how many minutes I want the app to stay awake. So, usually what I do is before I
-
20:19
, obre el vídeo en una pestanya nova
leave the house, I set the app to keep my laptop awake for a few hours, so that even if I'm sitting around watching my kid at summer camp or doing something else, I can actually continue to build and talk to Codex on my phone. Now, I have to admit that looking at all your Codex threads on your phone while you're out isn't really much better than browsing social media. So, I have to remind myself, and you guys should do this too, to basically go out and touch grass and just stay focused on hanging out with your family and friends because monitoring all your Codex threads and trying to get your agents to work all time is very addictive, and I'm not sure how healthy it actually is. Okay, so we covered the topics 1 to 6, but I wanted to save the best for last, which is how do you build beautiful websites for free with Codex and ChatGPT. So, going back to the app here, you can see that now Codex has a sites feature, right? And what this does, it it lets you publish any kind of website either to yourself and a few friends or publicly to everybody. So, in a previous video, I made a travel catalog for my upcoming trip to Japan. And I converted it into a site by just telling it to, "Hey, make this into a site." So, let's take a look at the site now. Let's open this here. And it's going to open in Codex's native app browser. And here is my Japan travel catalog. Again, it's got the torii gate at the top. And it's got all all the places that I want to go to down here. You can see how sharing this is just so much more engaging and interactive than sharing a list of bullet points. I can basically share this with my wife and kids to get them excited about the upcoming trip. Another site that I made that I want to show you guys is a safari math game to learn the multiplication tables, right? But so I'm trying to teach my kid the multiplication tables, and we basically live coded this game together. She's super into animals. So, basically the way the thing works is you answer the right question and it will show some confetti, and it will ask you more questions, and so on and so forth. And I can tell you guys that like she's much more interested in playing this game than just like learning multiplication from a standard math homework sheets. I think we built this app maybe a week ago, and by now she's memorized most of our multiplication tables, and I think it's thanks in huge part to this app. So, definitely build fun sites and apps with your kids using ChatGPT. It's actually incredibly simple. Just tell Codex and ChatGPT, "Hey, I want to build a fun animal website to learn the multiplication tables. Make it visually interesting." And go ahead and do it. And that's pretty much all you need with GPT-5.6 to build something like this. All right, let's look at one more site before we finish, which is this tastemaker plan. So, let's open it here.
-
23:14
, obre el vídeo en una pestanya nova
I've spent more than a decade working as a product leader, and during that time I've written a lot of PRDs and specs in Google Docs, right? And I've learned to keep them short because engineers don't like to read a lot of words. But, the reality is that people don't read them still. So, what's much more interesting, I think, is an interactive site they can click through to understand what the product requirements are. So, this is a new kind of PRD that I'm working on. This is a tastemaker website that lets you basically add movies, TV shows, video games, board games, and more to curate your taste and share with other people. So, this PRD has three different tabs. It's got the product requirements, the user problem, the goal, the solution, and the individual user stories. And it's also has the design. So, design, in this case, is the design principles, the style guide, and very importantly, the actual component library, the actual components that you should use. And what I found is that coming up with the style guide and component library first before you actually make a ton of mocks, it's extremely useful to keep all your designs consistent. And finally, the PRD also has the tech stack. So, what kind of tech we're going to use and why. The data schema that we're going to use, very important because it's very hard to change the data after you go public. And different trade-offs that we made. And perhaps most hilariously is both GPT-5.6 and cloud models keep saying that it's going to take like two or three weeks to build this stuff, when in reality it only takes five or 10 minutes to do it. So, I don't really know why it keeps trying to output, "Hey, it's going to take a long time." It's trying to lowball me in its estimates. I think at point I'm going to open source this new PRD format. Talking to your friends at OpenAI and Anthropic, a lot of them have basically switched over to using HTML for team review of plans instead of using markdown files or using Google Docs. And both companies are also building features like artifacts and sites to make reviewing and sharing and commenting on these HTML artifacts a lot more intuitive and engaging. So, I'm really excited for that. So, that was a bit of a digression. The point is make a lot of sites, share them with your friends and family, or make them public. It's probably the easiest way, in my opinion, to just get a web page online. And I should definitely try it out. All right, guys. So, we just covered a lot. I want you to remember seven action items from this video. Number one, if you haven't already, download the ChatGPT desktop app. If you're still using ChatGPT on the web, the desktop app is the biggest upgrade that you can do, because now ChatGPT and Codex can actually do stuff and help you get work done. Number two, if you're confused by all the different model options, which is extremely confusing,
-
26:06
, obre el vídeo en una pestanya nova
just stick with GPT 5.6 Soul with medium effort. As we saw from that chart, it gives you strong enough results without burning through all your tokens at once. Switch to high effort if you need to do heavier planning or maybe like a longer build. But generally, medium effort on Soul works perfectly fine. Number three, personalize your custom instructions, right? Tell ChatGPT how you want it to respond. Mine says to be concise, to avoid AI slop, to challenge me when needed, and explain technical concepts in plain language so that I can learn while I'm building. Number four, install plugins for the apps that you use every day. This is extremely important. Start with Gmail, Google Calendar, documents, presentations, spreadsheets, and Google Drive. And add tools like Slack and Granola if you use them. The more plugins you install, the more capable Codex and ChatGPT becomes. Okay, number five. There's different ways to organize your threads, but I think one of the first threads that you should start is a chief of staff threads that can help you manage your emails and your calendar. And we covered how you can use it to basically find action items and emails to unsubscribe from, and also plan for upcoming meetings, and even draft stuff in Google Docs to help you prep for a meeting. And most importantly, we talked about how to get Codex to do work proactively for you by just scheduling a bunch of tasks, for example, to run every week or even to run every day. Number six, we talked about using Codex from my phone. If I'm honest, I think about half of my Codex usage actually comes from my phone, and I rely on this app that I shared called Amphetamine, and again, I'll put it in the description, to keep my laptop on so that Codex can continue working. And finally, we talked about building and publishing sites, such as a travel guide, a learning game with your kid, or even a product plan. Stuff that's interactive and visually engaging that people actually want to read and consume. And you don't have to be a developer to learn all this stuff, you just have to ask ChatGPT or Codex to just build a website for you and describe it. And as usual, I'm going to include all the prompts that I used in this video in the first comment below. So, I hope you enjoyed this completely free tutorial. If you want to go deeper, I have all my best AI skills and courses on behindthecraft.com. So, check it out. I'm also building a complete Codex course with a lot more detail and my complete personal OS setup that I'll share later this summer. As always, I hope you appreciate no tutorials like this, and please like and subscribe if you enjoyed this video, and I'll see you next time.