Web3, Blockchain and the Metaverse | Metaverse Lead at Efiko | Incubation & Consulting

Google JUST released Gemini 3, the most powerful model in the world. 

And there's good reason it's breaking the internet...

I just one prompted an app & replicated the Financial Times' website in seconds. 

Here's what you need to know:

- Google say's its their next step 'on the path toward AGI'
- state-of-the-art reasoning
- deep multimodal understanding
- powerful vibe coding so you can go from prompt to app in one shot
- improved agentic capabilities, so it can get things done on your behalf, at your direction
- available to try for free now in Gemini app and Google AI Studio


I'd encourage you to push this model further than any other model to date. 

You can do things like: 

→ create full 3D games from a single prompt 
→ build complete platforms (UI + backend logic + AI workflows) from pasted ideas 
→ clone any UI from a screenshot or URL, add features by prompt 
→ create complex simulations and engines 


Here are just some of the best examples I've seen:

1/ full games such as 'Gemini Runner'
2/ created extreme physics simulations (AI ball-in-a-box test)
3/ creating interactive DNA
4/ a golden gate bridge simulation 
5/ a working simulation of a nuclear power plant

and so many more...


One of the best ways to get ahead is to understand what's possible. So I spent the last 12 hours stress testing the model

I've created a full Gemini 3.0 guide including 
→ all the model details you need to know
→ the best use cases now possible with Gemini 3.0
→ prompts to get top output and save time

Comment 'Gemini' below and I'll send it over

Elliot Garreffa

Co-founder @ Ghost Team | Scaling Companies with AI Growth Systems → ghostteam.ai | AI Agents & Automations | Ex-Capgemini Invent

We build an app to create all our proposals without leaving ChatGPT

And it now takes us less than 5 minutes per proposal 

Creating great proposals quickly is a big deal for a lot of teams out there, including us, because proposal creation is one of the biggest time sucks. Remembering everything said on calls, tailoring proposals in to each client, doing them all in our style and branding. 

We've been using n8n workflows with Gamma integrations for months but the design level needed just didn't meet our custom design requirements. And, we still needed to use multiple apps in the process.

This app and flow has made it 10x easier for us.

→ we attach the call transcript and any context 
→ prompt it to create the proposal
→ it creates it in seconds
→ the whole doc is perfectly brand aligned 
→ we can make any refinements and edits
→ just below, we have a full drafted email 
→ we can edit the email & send directly to the client, without leaving chatgpt
→ when they receive it, they can open up the live link 
→ they can even review and sign it 

It's great because the app is accessible to the whole of our team with our ChatGPT enterprise account. 

And we all move so much faster as we never even need to leave ChatGPT. We just have it as part of our existing workflows.


This really opened our eyes to the power of apps in ChatGPT. The use cases are endless. Think about how this could work with really any template: real estate brochure, legal contract drafting, sharing investor decks

Curious, what would you use it for?

NEW: Profound just launched something people interested in AEO need to pay attention to 

it's called the Profound Index - basically the first leaderboard for AI visibility

think of it like Google rankings, but for AI search


it's interesting because it's not keyword data. not hypothetical queries. actual prompts people type into ChatGPT every day.

the kind of insights you can get:
→ see where you brand ranks vs competitors across 12 industries 
→ shows you how keywords and prompts are performing week-on-week
→ understand which conversation clusters actually drive volume  


the critical difference to remember vs SEO? people prompt differently than they search

someone typing "what should i watch tonight" is asking for recommendations

someone typing "netflix vs hulu" wants a direct comparison

same topic. completely different intent. different brands will show up in each answer.


traditional SEO tools only track keywords - they miss this nuance

but Profound Index tracks real conversations, so you see the actual questions people ask and the intent behind them

this tells you exactly what kind of content to create for each conversation type


using tools like this gives you the baseline - where you stand, what's working in your niche, and which topics matter most

if you're serious about AEO, this is your starting point

link in the comments 👇

we've spent +30 hours building with the new ChatGPT apps SDK.

here's everything we learned on how to best build & distribute apps:


1. find your natural handoff point

the biggest learning: don't rebuild what ChatGPT already does. look for moments where users need your service.

we saw people drafting contracts, then leaving to post on Fiverr. that's why we built ChatGig. that context switch is your opportunity.


2. enable developer mode

you need ChatGPT Plus ($20/mo) to test anything.

settings → apps & connectors → advanced settings → enable developer mode


3. build your MCP server with widget resources

your MCP server needs to know how to send a widget (the UI) into the iFrame (the ChatGPT window)

use ngrok for local testing, deploy early to catch production issues

(FYI, we'll do a deep dive on the technical side soon)


4. write metadata that actually works

this is where we spent most of our time. ChatGPT reads your metadata to decide when to call your app, and most people get this wrong.

1. name format: domain.action (e.g., gig.create not createGig)

2. description structure:
- "use this when [specific scenario]"
- "do not use for [anti-patterns]"
examples for every parameter

the negative prompts prevent your app from being called at weird times. skip this and you'll get random invocations


5. create your golden prompt set

this became our testing framework. before you write code, define:
- direct prompts: "ChatGig, create a gig for this"
- indirect prompts: "help me find a freelancer"
- negative prompts: "what's the history of the gig economy"

write down expected behavior for each


6. test & iterate obsessively

we learned to track two metrics:
- recall: did ChatGPT call your app when it should?
- precision: did it only call when appropriate?

change ONE thing at a time. log everything: 
"changed description - recall 60%→75%"

keep a spreadsheet. no magic formula, just iteration based on real results


7. design for chat, not web

after building multiple apps, we learned what works:

good for ChatGPT: 
✓ fixed templates with clear outputs 
✓ workflow handoffs to external services 
✓ tasks under 2 minutes

bad for ChatGPT: 
✗ long-form content 
✗ complex multi-step workflows
✗ heavy visual browsing

if your workflow takes 10+ steps, this is probably not the best app use case


Apps SDK dropped October 2025. app directory coming "later this year."

once live, you have 800M+ users, zero install friction, and a 12-month window before the platform closes - the companies moving now are getting ahead


we've created a Full Guide on how to build apps in ChatGPT and you can access it for FREE

the guide covers everything: 
→ full video overview 
→ the opportunity (why zero-click discovery changes everything) 
→ the four phases of platform distribution 
→ what should you build? 
→ examples - apps we've built 
→ how to get your app discovered

links in the comments 👇

What questions are people asking ChatGPT about your company? And what pages are AI systems learning from to answer them? Most companies have no idea. We decided to find out.


When someone asks ChatGPT "What does Ghost Team do?", the answer comes from AI crawlers visiting our site and building their "memory" of our brand.

We built a tracking system to understand exactly what's happening. Over 2 weeks, we logged 333 AI crawler visits across 40 pages.

The patterns are wild - and they reveal exactly how AI systems are learning about businesses.


What we built:

A bot detection system that captures which AI systems visit, which pages they access, when they visit, and how they crawl.

Think of it as Google Analytics for AI bots.


A few things we're seeing:

1/ A homepage monopoly 
25.8% of ALL AI traffic goes to our homepage. ChatGPT-User spends 44.9% of its time there. This isn't true for every business, but if your early stage and don't have a huge amount of content, this is where AI forms its first impression. You've got to make sure this is content is a) very accessible / crawl-able by LLMs and b) has a clear value proposition so the LLMs don't misrepresent what you do.

2. Uncover hidden gems 
One of our detailed resources about building AI apps gets 10.2% of all traffic - second only to homepage. Long, technical content is a magnet for AI crawlers.

3. They come in bursts 
AI crawlers work in clusters, not random visits: 
→ GPTBot: 96% burst activity (crawls multiple pages rapidly) 
→ ChatGPT-User: 65.4% burst activity (user queries trigger exploration) 
→ Bingbot: 48.6% steady (consistent crawling)
When a bot lands on your site, you have a brief window (seconds) where it explores multiple pages. Your internal linking strategy determines how much the bot learns about you. We're exploring ways where we can guide them from our homepage to our most important content.


We're now in the era of AEO (Agent Engine Optimisation), not just SEO.

This is increasingly becoming the primary medium where users discover companies. As people shift from Google to AI assistants, your "AI reputation" affects customer acquisition.

The more you understand how these crawlers work, the more you can optimise your content to ensure AI crawlers actually find you and represent you accurately. 


We'll have more insights coming as we gather more data.

We wrote a full guide AI Crawler Behaviour for AEO sharing all our learnings in detail which is in the comments 👇

ChatGPT is now one of the biggest distribution platforms to ever exist.

Your business can now each over 800 million monthly active ChatGPT users. 


For the first time ever, you can meet the user exactly when they need you.

With the iPhone app store: 
- someone has a need
- they have to search a way to solve that need
- they go to the app store, try and find the best fit, download an app
- they have to go through a sign up process and onboard
- they can then transact 

With ChatGPT, this completely changes: 
- someone describes a pain
- your app surfaces to help them
- they can pay
Thats it...

No laborious searching, no sign up flows, no need to download an app 


Thats why every business needs to understand what's happening right now.

The distribution dynamics are shifting faster than most people realise. And the companies that understand this ecosystem first will have a 6-12 month advantage before it becomes saturated


------

In this week's Ghost Wire, we broke down:

We demo an app build: 
→ created Gradient Tweet MCP (turns tweets into visual assets) 
→ Real lesson: OAuth patterns that unlock cross-platform user data 
→ Most valuable apps connect existing tools - understand this now or be 6 months behind

Why the distribution model is different: 
→ App Store = high friction. ChatGPT = zero friction, 800M users already there
→ you're competing for relevance in existing workflows, not downloads 
→ the discovery game is algorithmic, not search-based

Discuss what OpenAI's Atlas browser means 
→ OpenAI launched their own browser on Chromium 
→ they're trying to own the operating system layer, not just build features 
→ control the browser = control discovery, distribution, and monetisation

Technical barriers being lower (and dangerous): 
→ any dev can build ChatGPT apps - thousands flooding in 
→ your advantage isn't tech - it's understanding distribution mechanics

Why niche focus wins: 
→ OpenAI hiring experts across verticals (finance etc.) 
→ they're analysing their top API callers. this will become their product roadmap
→ you cannot compete on breadth 
→ only path: impossibly deep in specific workflows they'll ignore

------


Watch the full breakdown (link in comments). We also share a Full Guide on everything you need to know.


Ghost Team is build ChatGPT apps for companies that want to get ahead. Check out our website to learn more.

Like with websites and mobile apps, the ones that move first and the ones that win.

AI slop is everywhere. And most of us are sick of it. I'm afraid to tell you, it's only going to get worse. That's why, we're making a bet on something different: 

Authentic conversation. Let me explain...


Over the next 12 months, AI content will flood every single platform, even more than it has today. Short-form and copy heavily platforms will be worse: TT, IG, LN, X. 

I also believe AI content will get so good, it will become impossible to filter. People will spend more time curating their feeds, figuring out who they trust and finding parts of the internet where they can find genuine connection.

People will desperately want authentic conversations with real experts in their field.

This will manifest itself in different ways... watching more long-form content, engaging in live streams, going to in-person events. In short, anything that achieves the following: authentic and real human connection


At Ghost Team, we're thinking about each of these developments. But one direction we believe will be more important than others: Community 

Community has been important for a long-time. But soon it'll become more important than ever. When the internet becomes more divided, having a small, tight-knit community of people around a common interest will be something everyone seeks. 


Over the last 4 months, that what we've been building quietly at Ghost Team: a highly curated network where people serious about AI, there to help each other win:
→ incredibly high talent density (a no. of +$1m ai startups, Google, Boston Consulting Group (BCG), Workday, Amplitude, Capgemini Invent to name a few) 
→ access to real working systems (mcps & apps in chatgpt, claude setups, n8n workflows) 
→ proven playbooks and learnings from people scaling AI right now 
→ a place people can go to ask questions and learn from top talent 
→ even access to events! just a few days ago, we collaborated with a community member [ T-Bot ] 👾 to give everyone in the community access to his Voice AI Conference in Paris this December (would highly recommend people checking it out)


Some feedback from a few community members:

“With the world moving so fast you can’t just rely on receiving information or tutorials , you need to connect with others who are actually building and experimenting and this is exactly what I have found with Ghost team, those who are experimenting and uplifting each other to push ourselves - I’m very proud to be apart of ghost community" - 🐨 Brenton Price

“Being part of Ghost Team keeps me surrounded by builders who are always two steps ahead - experimenting, sharing, and inspiring. It’s the perfect environment to stay at the edge of AI, which is exactly what I need to serve clients in my consulting business" - Moritz Schuster


And the best part, the community is completely FREE. 


If you're a top founder, executive or professional looking for a place to get ahead in AI, Agents & Automation, join us! Apply in the comments, takes a few mins 👇

OpenAI just confirmed the biggest distribution play since the iPhone App Store.
Their CEO of Applications YESTERDAY revealed what's coming - here's what you need to know:

Fidji Simo (ex-Meta, ex-Instacart) now runs all Applications at OpenAI - ChatGPT, monetisation, future distribution. She just did interview with WIRED.  


→ Ads coming to ChatGPT free tier
Not live yet, but coming. Commerce-focused. It'll be a new format, not Meta's model. Think promoted agents, recommended workflows, sponsored tools. MAJOR traffic source.

→ Compute bottlenecks are real
Many features ready but can't launch. GPU constraints blocking rollout. Example: Pulse (proactive assistant that reads calendars, monitors for you) is Pro-only because they lack compute for free tier.
High-compute features debut Pro/Enterprise first. As compute scales, expect these to open to broader distribution and potentially 3rd-party integrations.

→ Proprietary chip + hardware coming
It's confirmed, they're building their own chip. "Mysterious hardware product" referenced...

→ Sora as distribution channel
Video quality improving. Strong guardrails. Copyright holders want monetisation partnerships, less see if they can limit the legal battles here.
Future op: API or creative-agent distribution inside Sora.

→ Jobs displacement + marketplace = upskilling opportunity
"Jobs were one of my first two initiatives" - she sees major disruption coming to certain categories. In response, OpenAI's is certifying 10M workers as "AI-ready" and launching a jobs marketplace to connect them with opportunities.
The opportunity: Upskilling apps and training agents will have massive demand AND a distribution channel through OpenAI's jobs platform.

→ OpenAI will build valuable verticals themselves
I quote: "If we build that" - referencing the high adoption areas such as personal shopping, travel, finance advisor, health coach etc. They need the revenue and will go after the obvious use cases. 
However, they do believe in "agents for every industry", "either by us or by enabling third parties"

→ "Reinvent every category of software" - your opportunity
OpenAI can't build everything. Third-party apps fill the gaps. Find categories OpenAI won't prioritize. Build deep vertical value where they won't compete.
Avoid personal shopping, travel, finance, health - likely OpenAI targets.
Find the long tail they'll leave to the ecosystem.


And one more thing - take this with a pinch of salt.

Notice the "compute is the constraint" narrative? Microsoft just said the same thing. Convenient messaging to justify +1$ trillion in spend commitments.

And yes, they want you building on their platform. More developers = strong ecosystem = winning platform. But platform risk is real. What happens when OpenAI builds exactly what you've already built?


Apps & ads alone on ChatGPT are some of the greatest distribution opportunities we've seen since the iPhone app store, for those that get it right.

Good luck & happy building.

The world's best agentic model is now Chinese and open source 

Kimi k2 thinking BEATS gpt-5 AND claude sonnet 4.5 on most benchmarks

And, it's a 1/4 of the cost 


Here are the numbers:
→ 44.9% on humanity's last exam (beats GPT-5 and grok-4) 
→ 60.2% on browsecomp (2x human baseline)
→ 71.3% on SWE-bench verified (massive gains in coding)


So, what's it best for?

Agentic coding: 
→ builds and debugs complex codebases across hundreds of sequential tool calls 
→ plans, verifies, and revises code without human intervention 
→ handles long-chain reasoning for software development

Agentic search: 
→ performs goal-directed web research with 200-300 tool call chains 
→ gathers facts, analyzes data, and synthesizes findings automatically 
→ more than 2x better than humans at complex browsing tasks

Long-context work: 
→ 256K context window means you can load entire documents, extended conversations, or massive tool outputs all at once 

Complex reasoning: 
→ test-time scaling means it thinks harder on difficult problems 
→ interleaved thinking between actions keeps context across hundreds of steps
→ PhD-level problem solving through deep reasoning chains


All of this and to repeat, it's 4x cheaper than GPT-5 and claude sonnet 4.5...

And completely open source.


Jensen Nuang (Nvidia CEO) said just last week that the US is at risk of losing the AI race. He's right - China is doing all this with less powerful GPUs.

There is more pressure than ever now on the US to keep pushing the frontier models to stay ahead and evolve their open source ecosystem.

Google drops Gemini 3 soon... all eyes on them on them to see if they can match Kimi's numbers.


So the question is: if you can get frontier-level agentic AI for 25% of the cost, what are you choosing?

So... I just built an app that takes creating Client Proposals from 2 hrs to 10 mins

All without leaving ChatGPT...


Here's the problem: proposals take hours to create manually 

→ valuable time is wasted formatting and branding 
→ you lose momentum between the call and sending means losing leads
→ if you're traveling without your laptop, you're stuck

When this is scale this time cost across your whole team it becomes very costly.

Companies spend thousands of dollars in sunk time which could be spent on higher value tasks.



So, we built Proposal Generator inside ChatGPT.

Here's how it works:

You finish a client call and open ChatGPT and give it the call transcript...

1. It creates a proposal immediately 
→ prompt it: "Create a proposal for [Client Name] - $50K website project, 8-week timeline" 
→ ChatGPT opens up the Proposal Generator app and creates the proposal in seconds

2. You can refine it in the chat 
→ you edit the copy, adjust pricing, customise sections 
→ it's fully tailored to YOUR brand, YOUR style

3. You can send it directly to the client 
→ once it's ready, you can click "Send Proposal" 
→ a live link to your proposal lands in client's inbox with a fully writtenemail 

This happens all inside your existing workflows.
You've never had to leave ChatGPT.


This means:
✅ professional proposals created in minutes 
✅ sent directly to clients via email 
✅ works from your laptop and your phone 
✅ fully customisable to YOUR brand and style


This saves hours per week. Not just for you - your whole team can use the same application as an internal tool, with your brand guidelines built in.

----
💬 We're giving people early access and building this custom for a few brands.
If you'd like to join the waitlist, comment 'proposal' below and I'll send you a link to sign up.

♻️ If you know someone who may find this proposal generator useful - please reshare this
----

We believe Apps in ChatGPT will be central to the future of work. 

I'd love to hear - what proposals or template documents would your team generate with this tool?

AI is going banana's once again...

You can now make Infographics with Google news model Nano Banana Pro! 


1st Image: 
- you may think its the Golden Gate bridge, but it's actually 25 de Abril Bridge in Lisbon
- its easy to create simple technical annotations on an image 

2nd Image:
- I gave Nano Banana Pro one of my LinkedIn posts on an proposal generator app we made in ChatGPT
- I asked it to create a cartoon on the steps and it did it perfectly 

This may look simple, but this was not possible at all with previous models. things like spelling, getting accurate images based on the description you gave, being able to provide a lot of design direction were never possible before


A few tips from using the model so far:

- more detail = better results – the more specific your prompt, the better the output

- iterative refinement works now – unlike previous models, you can ask for precise updates after the first generation. Just be as specific as possible with your refinement requests

- use Gemini as a prompt writer – ask Gemini to create the prompt first, then feed that prompt into Nano Banana Pro for better results

- if you found a prompt that is creating good images, replicate the prompt structure:
1. ask Gemini to create a new prompt in that same style
2. use that prompt to generate your image

- if you want to match a specific design, upload a reference image to Gemini and ask it to create a prompt that replicates that infographic style


Full prompts I used are in the comments (to long to share in this post) 👇

OpenAI launched Group Chats.
Multiple people + ChatGPT in one shared conversation. Everyone sees the same context. No more screenshot hell.

At Ghost Team we speak to so many teams that all face the same problem: when you build something in ChatGPT, you hit a wall the moment you need collaboration.

→ you can't bring your team into the conversation 
→ context gets lost in Slack threads and screenshots 
→ you can share projects but then there is no way to collaborate 
→ so what happens? you work in isolation, then manual sync


Group Chats will change that.

How it works:
→ one person creates a group link
→ others join via the link
→ ChatGPT participates alongside everyone else
→ everyone sees the full conversation history
→ anyone can ask follow-ups or add context


Think about what this enables. 

Not only can you now collaborate with internal teams. You can extend your AI workflows with anyone you're working with...


Building a product with a freelance developer? Invite them into the ChatGPT conversation where you've already defined requirements.

Created a first draft of a legal doc? Ask your lawyer to join the chat so you can directly discuss and update the draft.

Running a client project? Bring them into the strategy conversation instead of summarising afterwards.


It makes you think about the potential impact to collaboration platforms.. Slack, Teams etc. They're not going anywhere for a while, but teams will definitely migrate some discussion directly to ChatGPT.


Right now, this has rolled out to Japan, New Zealand, South Korea, and Taiwan for Free, Plus, and Team users. Soon, it'll be global.

Not long now till everyone sees the full context. Everyone can ask follow-ups. Everyone stays on the same page.

here's how I optimised 20+ resources for AEO using Cursor & Claude Skills:

it turned 1+ hour of manual work into 5 minutes


if you've got a backlog of blogs or content that hasn't been optimised for AEO, you're sitting on an opportunity to make this optimised for LLMs

this process is the quickest way you can convert it

i created a claude skill that cleans docs, preserves hyperlinks, and optimises for LLM search (query-answer alignment, dense facts, authority markers)

then used cursor to batch process all files automatically - it just iterates through each one and applies the skill


if I hadn't of taken this approach, I would have still been able to use Claude Skills but in Claude Desktop which means I couldn't have got it to run programatically, therefore taking me much longer 


if you understand how cursor + claude skills work together, you can automate almost any repetitive content workflow

full prompt I used in the comments 👇

AI bots are crawling your website every day - ChatGPT, Gemini, Claude - and most companies have no idea. So we built a system that tracks AI crawler behavior & an n8n workflow that automates weekly AEO insights. 

Sharing full guide on how to set this up below 


--
Want to grow smart with AI Agents & Automations? 
Check out → ghostteam .ai

So... we built a system to learn how AI bots are learning about your company 

You can now get AI-bot crawler insights delivered to Slack every week


Every company is getting their website crawled by LLMs - ChatGPT, Gemini, Claude etc. - and most have no idea this is happening, let alone what it means.

But this data is gold.

Understanding what these crawlers are doing tells you what LLMs are looking to know about your company. With this, you can reverse engineer exactly what you need to do to improve your AEO and get more visibility in AI responses.


We've had this tracking system set up for a while, but manually reviewing the data was a nightmare.

So we automated it.

Every Monday, an n8n workflow: 
→ Pulls the latest crawler data from our database 
→ Filters for significant activity patterns 
→ AI analyzes trends and anomalies 
→ Generates a formatted report 
→ Sends it directly to Slack (or Telegram)


Now we see insights like:

Traffic concentration: 
→ Which pages AI is obsessing over 
→ Where ChatGPT-User spends most time 
→ Hidden content that's attracting crawler attention - one of our detailed resources about building AI apps gets 10.2% of all traffic

Crawl patterns: 
→ GPTBot burst activity 
→ When bots are most active 
→ Which pages they explore after landing 

Strategic gaps: 
→ Important pages being ignored 
→ Internal linking opportunities 
→ Content that needs optimisation


You can't optimise for AEO without understanding how AI systems are learning about your brand.

Understanding how AI-bot crawlers are visiting your website gives you the insight you need to become visible to LLMs.


We're giving away this set up for free: 
- full guide on setting up AI bot crawlers 
- n8n JSON file automating insights 

Link is below in the comments 👇 

Share this if you a team who needs this set up ♻️ 


--
Want to grow smart with AI Agents & Automations? 
Check out → ghostteam .ai

If you're building on OpenAI's platform without an exit strategy, you're making a BIG mistake...

The same mistake Zynga made with Facebook in 2012 that crippled them


Here's what just happened: OpenAI launched their App Store

Combined with their API and AI cloud infrastructure (coming soon), this is the biggest platform opportunity since 2008

And it is a textbook platform play OpenAI is executing 


These platform opportunities are incredibly rarely: 
→ Microsoft 
→ Facebook 
→ Google SEO/Ads
→ iPhone App Store 
→ OpenAI & ChatGPT (NOW)

In all of these, we've seen the same 4 phases play out.


Phase 1: Market conditions met
- OpenAI (800M users), Anthropic, Google, xAI, Meta competing for LLM dominance

Phase 2: Moat identified
- OpenAI's advantage: context + memory. the system that knows you and compounds with usage. They're running away with the audience

Phase 3: Platform opens ← WE ARE HERE
- API live. Apps SDK dropped. AI Cloud infrastructure coming. Platform subsidizes growth. Apps will grow fast as they're handed discovery

Phase 4: Platform closes (coming 2026/27+) 
- Absorb top use cases. Reduce organic distribution. Force paid placements

It's Phase 4 where you need to be extremely careful 


Remember what happened on previous platform plays? 
- Microsoft: Let Lotus thrive → built Excel. Let WordPerfect win → built Word
- Facebook: Partnered with Zynga → then competed

OpenAI is following the same playbook: 
- Open platform, invite developers 
- Study usage (token/app usage = their roadmap) 
- Build native features for successful use cases 
- Reduce third-party reach, own revenue

Right now, OpenAI is studying top token users: Duolingo, Notion, Shopify, HubSpot

Why? They HAVE to

They need high-value use cases

$1.4 trillion infrastructure can't survive on tokens that drop 90% in price yearly


So, what do you do as a company?

I believe, most companies actually have no choice but to play the game

For your customers, having an app in ChatGPT will soon be an EXPECTATION

As workflows move inside LLMs, if your brand doesn't have an app, users will choose competitors who do

It's a game you HAVE to play. You just need to be cautious of HOW to play it


My advice: build distribution outside ChatGPT from day one
→ Own customer relationship (email list) 
→ Standalone web version 
→ Social following independent of platform 
→ IP they can't easily replicate 
→ REALLY niche down


What are your next steps? Depends on your business stage:

Incumbents: 
→ Allocate experimentation budget now 
→ Build early, optimize over time 
→ Your competitors will build, you can't afford not to

Startups: 
→ All-in or stay out completely 
→ You don't have resources for multiple bets 
→ Pick one platform, go deep, move fast


This is a big deal. A huge decision every company needs to make over the coming months

We do a full deep dive into this topic in Ghost Team Weekly

You can find the link in the comments

Elon Must just said apps and operating systems will disappear in 5-6 years

you'll just talk to AI and it'll generate everything in real-time


it might sound crazy. but what I think he's saying indirectly is that it won't be a screen for too long

it'll be a device like Neuralink where we have a brain computer interface. you'll live your life, potentially without screens, orchestrated by AI... crazy to think and sounds very like the matrix 


however, we're already starting to see the world move in the direction he describes...

look at openai. chatgpt has 800 million monthly active users. they just announced an app store and soon, AI cloud infrastructure - they're building the platform for businesses to develop on top of chatgpt

as the app ecosystem grows, people will barely need to leave the interface. it understands your intent and pulls in whatever app is needed right into the chat

so, do you need apps? yes. but where do you access them? not through your phone's operating system - directly through your AI chatbot


however, while AI can orchestrate and generate a lot, i don't believe it will necessarily generate everything like Elon says

if AI generates your linkedin feed from scratch every time, you get inconsistency. your version looks different from mine. and unless AI is hardwired into your brain, it can't perfectly understand your intent every time 

what you actually want is AI as the access layer - fetching real content, loading real interfaces, connecting to real platforms


so the shift isn't apps disappearing. it's WHERE they live

apps move from your operating system into your AI interface. they become services AI orchestrates rather than icons you tap. you still see the same linkedin UI as everyone else, but access it through AI instead of iOS

real-time news, social feeds, platform interfaces - these still exist as stable experiences. AI just becomes the smart way to navigate them


companies are starting to develop apps NOW for chatgpt. there is a window of opportunity to get first mover advantage, giving you distribution to their +800 million users

we created a full guide on building apps in chatgpt which I'll put below in the comments. worth checking out if you're looking to get ahead on the opportunity 


what do you think - will apps truly vanish or just become integrated into the AI experience we know today?

Gemini 3 Pro is about to break the internet. 

Better on every single benchmark. 

Expected to be released publicly later today...

We built an app to find freelancers without ever leaving ChatGPT

And it now takes us less than 2 minutes to go from first draft to posting a gig

We kept running into the same problem.. we'd create a first draft in ChatGPT - a legal doc, design brief, content outline - and then we'd have to switch to Fiverr to find someone to finalise it.

Copy context, rewrite the brief, explain everything again - it was clunky and time consuming as it was broken up into two separate workflows.


So we built ChatGig - a freelance marketplace that lives directly INSIDE ChatGPT (using the new OpenAI apps SDK).

→ you create a first-draft doc or a brief in ChatGPT 
→ prompt it to "use ChatGig to create a freelance gig for this" 
→ it auto-generates the gig based on our conversation context 
→ summary, success criteria, and required skills are already filled in 
→ you can refine and edit anything 
→ add in the timeline and budget you have
→ save the gig and share it with freelancers directly 
All this, without every leaving ChatGPT.

Right now, this is more of a concept app as we don't have the freelancer side of the marketplace but it'd be very easy to distribute these gig's to freelancers in each niche. 


This kind of experience makes so much sense for companies or individuals looking for freelancers because you never have to leave your workflow.

And for freelancers, it's great because you're already receiving the first draft or brief document they've already made in ChatGPT meaning you can see their thought process and are up to speed more quickly. 


This is another eye opener around the power of apps in ChatGPT. The use cases for working with freelancers in chatgpt are endless: 
> legal contract finalising 
> sharing draft logo designs with designers 
> finding engineers to execute a product requirements document
Just to name a few.

Now with group chats in ChatGPT, you could even see a workflow where freelancers are brought in via a gig into the conversation to carry out specific pieces of work...


This is the future - LLMs become the main interface for more and more of our work.

Curious, what would you use ChatGig for?

Agencies, B2B Sales Teams, Professional Services, you can now create proposals that used to take 2 hours in 10 minutes, without even leaving ChatGPT


Most companies are still building proposals in Google Docs or PowerPoint, manually creating them and waiting hours formatting


Now, you can use an app inside ChatGPT to make proposals in minutes: 


1/ upload your transcript & context

Start by uploading your client call transcript or meeting notes

The AI processes everything to extract key info and context

No manual note-taking, no trying to remember what they said


2/ tag the app & prompt

Tag @ProposalGenerator inside ChatGPT

Prompt it to generate the proposal, with any additional detail on what you'd like: "Create a proposal for [Client Name] based on this call transcript and context." 


3/ create proposal

Watch as it generates a professional proposal with your brand template

Custom branding, colors, styling applied instantly


4/ review and edit

Make adjustments, fine-tune content, all without leaving ChatGPT

No more jumping between 5 different tools


5/ send to client

Send your polished proposal directly to your client with a single click

They receive a fully drafted email with a link to a beautifully branded proposal 


All of this without needing to leave ChatGPT and done within 10 minutes.


The results?
- clients get wowed by professional proposals with lightning-fast turnaround 
- you move from first call to signed contract 3-5x faster 
- your team stops spending 40+ hours/month on proposal writing 
- beautifully branded documents every single time


If you or your team would like the tool, we're collecting interest: 
→ comment 'Proposal' 
→ I'll DM you the details

taplio

Elliot Garreffa's Linkedin Analytics

Elliot G.

Performance ScoreElliot G.'s LinkedIn Performance

Elliot G.'s Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Elliot Garreffa's Linkedin Analytics

Elliot G.

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Elliot G.'s LinkedIn Performance

Elliot G.'s Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Elliot Garreffa's Linkedin Analytics

Performance ScoreElliot G.'s LinkedIn Performance

Elliot G.'s Best Posts (last 30 days)