Siemens

Focus: Digital Services, Metaverse and enabling technologies (incl. AR/VR, AI)
Digitalization community

Digital Innovation and Product Portfolio Management | Siemens Mobility

Dear Ava

Dear Ava is a curated online store and community for unique, personalizable gifts & packaging

Founder

Siemens Advanta Consulting

Focus: IoT, AI, blockchain use cases
Digitalization & Innovation community

Senior Consultant - Digital

Consultant - Digital

Consultant - Strategy

ConsciousCoders

ThinkTech e.V. (formerly: ConsciousCoders) is an independent nonprofit association of critical thinkers and makers, researchers and entrepreneurs, coders, philosophers, lawyers and artists. 

We are driven by the goal to understand, discuss, and shape the impact of AI and other digital technologies on individuals and society.

Co-Founder

Siemens Management Consulting

Internship | Strategy Consultant

VINCI Energies

Internship | Analyst in Strategy & Innovation

Internship | Power Systems Engineer

innogy SE

Internship | Portfolio Analyst

Internship | Circuit Design Engineer

I gained experience in:
📊 management consulting (large scale digital transformations, G2M strategies), 
💻 product and portfolio management (AI, AR/VR and IoT), 
💎 innovation management (use case and business model design), 
🤝 community building & management (corporate, web2 & web3), 
all while launching and scaling my own e-commerce business 🛒

Anna Erlenkamp

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

Elliot G.

Finally taking a moment to reflect on how much fun the last few weeks of work have been.

I had the pleasure of interviewing Roman, Co-founder and Chief Business Officer of Nebius.

Then flew to Hamburg to meeting Ingo and the Statista team to present to them on ChatGPT & MCP Apps.

From there, Vincent & I went to Paris where I give a talk on ChatGPT & MCP App discoverability - we connected with a ton of amazing people. Thanks again to the Alpic team for hosting us. 

Then back to Lisbon and connected with the Paygentic team. Great to meet you Aine, Susan & Samuel.

Many chats with Hanh, helping companies build and prototype ChatGPT apps with no code with usefractal .dev

Flew to London to do a full-day AI acceleration session with Roslyn and the Noetic team.

Great chatting with Ruth and Gwenaelle from ManpowerGroup about the Intent-Based Web while I was there.

And then we got the news:

We'll be speaking at MCP Dev Summit in New York on April 2nd.

Two talks:

1. MCP & ChatGPT App Discoverability (the hot topic right now)
2. Design Principles for Building World-Class ChatGPT & MCP Apps

If you’re in New York at that time, let me know and we can grab a coffee!


Grateful for the people, the conversations, and doing some of the most exciting work in AI.

Full speed ahead!

Elliot Garreffa

Get Your ChatGPT App & MCP-App Built & Discovered | Co-founder, ghostteam.ai & appdiscoverability.com

A major announcement from Shopify to their merchants:

ChatGPT, Gemini & Copilot can now sell your products for you. Launching by default for every merchant.


Here's what makes this interesting:

Last week, OpenAI quietly killed "Instant Checkout" - their attempt to handle purchases directly inside ChatGPT. Very low conversions. Users researched products but didn't buy. They never even built a sales tax system.

OpenAI's exact words: "Instant checkout is transitioning to apps, where purchases can occur more seamlessly."

So transactions are moving to ChatGPT Apps - the branded apps that companies build inside ChatGPT. 

And days later, Shopify announces "Agentic Storefronts," where every Shopify store gets plugged into ChatGPT, Gemini, and Copilot by default - with checkout handled by Shopify's infrastructure.

It looks like OpenAI tried to own the full transaction and couldn't make it work. 

Now they're pushing it to the apps layer - and Shopify is positioning itself as the infrastructure behind some of those apps.


I think its clear agentic commerce isn't dead but it's taking time to find it's final form. 

We're moving into the next state which is splitting commerce into two layers: AI handles discovery and existing commerce platforms handle the actual sale.


Full release on agentic storefronts below 👇

The Intent-Based Web is coming.

A few weeks ago I presented this thesis at Statista HQ in Hamburg - invited by Ingo Schellhammer, Statista’s CTO, to share insights with their product and tech teams.

The core idea: discovery is shifting from search bars and app stores to conversations.

We explored:

→ Why "searching for apps" feels increasingly broken
→ How conversational AI changes the discovery layer
→ How to design applications for the intent-based web

I've been impressed with what Statista is already doing in this space - having their MCP available in Perplexity, Copilot, Langdock and soon to be expanding to other LLMs.

Thank you to Ingo and the Statista team for having me - it was great meeting everyone. A really sharp group of people with lots of great questions, deep product thinking, and a lot of excitement for what's coming next.

This is why you build MCP Apps

Amplitude getting a 2x boost in retention since MCP UI was released in Claude 

Moral of the story...

Meet users where they are: using agents in AI Chatbots and IDEs

cc Ferruccio Balestreri

A major announcement from Microsoft stating they're adding support for MCP Apps in Copilot. 
This gives those with an MCP App access to 400 million enterprise users

We're starting to see the MCP App standard get adopted across the board. This means that with one app, you can reach all these audiences and more: 

→ ChatGPT (ChatGPT Apps) - 900M weekly active users
→ Microsoft 365 Copilot (MCP Apps) - 400M+ paid seats
→ VS Code (Plug-ins) - 15M+ monthly active users
→ Cursor (Plug-ins) - 1M+ developers
→ Claude (Claude Connectors) - 18.9M monthly active users

One protocol. Every major platform.

MCP Apps are becoming a clear path to AI discoverability. 

Full news release from Microsoft in the comments 👇 

---
If you want to stay ahead of MCP Apps and get what apps are launching across ChatGPT, Claude, Cursor and now Microsoft Copilot straight to your inbox, sign up at https://lnkd.in/ejnAZGD3

claude can now do this...

directly inside your conversation it can build you interactive charts, diagrams, and visual experiences 

you ask a question and claude decides whether a visual would explain it better - then it just builds one for you without needing any code or specific formatting prompts from you

artifacts were the first iteration, now this is the level up.


but here's the cool part: this is an MCP App under the hood

the same MCP standard that powers claude connectors, chatgpt apps, copilot apps, cursor and vs code plug-ins

this is what MCP apps actually unlock. rich, interactive experiences inside AI conversations that go way beyond text responses and code output


Drew Bent from anthropic said it best: "it's time to go beyond the chat interface in AI and just text"

the visual web for AI is being built right now and it's being built on MCP and MCP Apps

Hanh Nguyen and I have collectively we've built well over +100 ChatGPT Apps.

We jumped on a call to breakdown every insight we've learnt on how to build world class ChatGPT & MCP Apps.


Here's a high-level breakdown of some of the things we discussed: 

1. Common confusion: ChatGPT Apps are not GPTs. Not web apps. Not mobile apps. They're a completely new paradigm.

2. We're moving from the "fragmented web" to the "intent-based web." You use to design for you finding the tool. Now you design for the the tool finding you.

3. Don't start by listing features and building tools. Start with use cases. What will people actually type?

5. The model controls the flow, not you. Biggest mental shift from web/mobile dev.

6. "Tool description" is the wrong name. It's a tool instruction. Treat it that way. Most builders write "Search hotels" and wonder why it never gets called. Be specific.

7. Include WHEN to use your tool AND when NOT to. Negative use cases matter just as much.

8. Add example prompts to instructions. "Use when people say X, Y, Z" dramatically improves call rates.

9. Leverage ChatGPT's intelligence inside metadata. e.g. "If user doesn't specify a city, check history and suggest favourites."

10. Prototype with mock data before building any API if you don't have one already. Test the conversation until it feels right. Only then build backend.

11. Your existing APIs won't work necerssarily work as well as you expect. You'll likely need to redesign for conversational use cases.

12. Think bulk operations. Instacart adds all recipe ingredients in one call - not 20 separate ones.

13. Know partial tool calls vs follow-up messages. Same-UI refinement = partial call. New results = follow-up. Get this wrong and the experience feels broken.

14. ChatGPT knows more about users than almost any platform ever. Use that for personalisation.

15. Two surfaces to optimise: the UI widget AND the chat. Websites have UI. Chatbots have chat. You have both.

16. Use developer mode to inspect how top apps handle tool calls. You'll learn a lot from their metadata.

17. Build retry logic into tool instructions. Failed searches are inevitable - define how to recover.

18. ChatGPT App Store discovery is one of the key ways your app gets found. Optimise your metadata, keywords, and descriptions from day one and track how you compare against competitors. Lots of app store data accessible at appdiscoverability .com/track.

19. Organic discovery is why most people are excited about building ChatGPT apps. Your app gets recommended at the end of a user's prompt based on their intent. It's not fully live yet - but there's a lot you can do now to optimise for when it launches.


This is just a high-level snapshot. We go deep on every single one of these in more detail in the full conversation.

Link in comments 👇

If the LLM can't find you, you don't exist.

That's the reality of where this is all heading.

We're shifting from "you find the tool" to "the tool finds you."

That's the Intent-Based Web - and it changes everything about how brands get discovered.


I spoke at MCP Connect in Paris last month about exactly this.

Everyone's focused on building ChatGPT Apps & Claude Connectors. Almost nobody is thinking about what happens after you ship - how does your app actually get discovered?


Here's the framework we use at Ghost Team - 5 components that work as a continuous loop:

3 foundations to build:

→ Be Findable - get into the registries (MCP official, GitHub, Smithery, ChatGPT Store, Claude Marketplace). If you're not listed, you don't exist.

→ Deliver Value - optimise your tool metadata, descriptions, your golden prompts, test this at scale. The LLM needs to know exactly when to surface your app.

→ Be Loved - high engagement, high retention, strong ratings. Organic discoverability will be driven by signals that your app is one people come back to.


2 outcomes to measure:

→ Static Discovery - are you ranking for the keywords that matter in the app stores?

→ Organic Discovery - when someone types a relevant prompt into ChatGPT or Claude, does your app appear? (coming soon)


The loop: you build the foundations → you measure the outcomes → the data tells you what to improve → you go back and optimise the foundations.


Full talk linked below which much more detail on the above framework - huge thanks to Pierre-Louis, Anastasia and the rest of the Alpic team for the event and the opportunity to speak!


Right now, the main ways to get discovered right now is in the ChatGPT App Store. We track all the data on the App Store and have also created a Score for how good an apps App Store visibility is. Check out each companies score here:  appdiscoverability .com/score

If discoverability is something you're thinking about for your app, feel free to reach out me directly.

Travel just had its biggest year since the pandemic.

And now it's one of the hottest categories in ChatGPT's App Store.

With 16 apps in the category (data from appdiscoverability .com/track)


Why travel and hospitality?

It's consumer-focused. High intent. High volume.

But there's something bigger happening:

Consumer habits are changing.

People aren't just Googling "best hotels in Tokyo" anymore.

They're asking chatbots and AI agents to help plan their trips - or even decide for them.

The travel industry knows this. And they're moving fast.

Now, you're already starting to see the category in the ChatGPT App Store becoming "competitive" with 16 apps focused on travel that have been launched.

But it's not. Here's why:

There are billions of travel prompts people type every day.
"Best hotels in Tokyo for families"
"Weekend trip from London under £500"
"Romantic getaway with spa and beach"

Each prompt is a unique moment of intent.

And right now? Maybe 0.001% of those prompts have a ChatGPT App built for them.

The travel category alone could support thousands of apps.

Because the nuance is endless:
→ Solo travel vs family travel vs couples
→ Budget vs luxury vs mid-range
→ Adventure vs relaxation vs cultural
→ Weekend trips vs month-long sabbaticals
→ Business travel vs digital nomad setups

Every combination is a different user. A different prompt. A different app.

Congrats to Awaze (Cottages App), Jinko, Voyage Privé, Accor, Hyatt,  MakeMyTrip, Viator, Priceline, CHECK24 Vergleichsportal GmbH and every travel app that's available for people to try now.

We're at the beginning of the intent-based web for travel.

Claude Cowork completely flips on it's head how you can do SEO & AEO. Not "AI helps you write faster." The entire workflow - research, strategy, content - done in one session. Here's how you do it: 

Anyone that's done SEO work knows a lot of the heavy lifting has always been the workflow before the work.

You need keyword data. Competitor analysis. Cluster mapping. Content briefs. Then you need to write. Then edit. Then format.

These kind of workflows if you're doing them with a traditional agency, to create 3-5 SEO optimised blogs, can take weeks. 


And today? I'm not exaggerating when I say this but end-to-end research, strategy and content can be done in less than 20 minutes. EVEN if you don't know how to code.

How? Claude Cowork 


Firstly, what it is Claude Cowork? Put simply, its a nice interface on top of Claude Code (arguably the worlds most powerful coding agent) that runs directly on your computer - with skills, connectors (MCP Apps), and sub-agents and commands that let it complete full multi-step workflows autonomously.

Here's the SEO & AEO content workflow I ran:
→ Connected GA4, GSC and Ahrefs into Cowork 
→ Triggered a keyword research skill (it has the full methodology baked in) 
→ It mapped clusters, scored opportunities, identified low-difficulty wins 
→ Built a 20-article content roadmap - ranked by priority 
→ Then immediately started writing the first 5 articles using the SEO content creation skill 
→ Each article came out with full metadata, FAQs, internal link suggestions

Doing this exact workflow was simply not possible before unless you used claude code in the terminal. You try and do this in Claude and you'd hit context window limits and end up copy-pasting between tools.

Now you do this all from a single prompts as long as you have the right connectors and skills set up.


I actually have this workflow set up in my Openclaw which is connected directly to my CMS so can publish automatically as well, but more on openclaw workflows next week...  

In the comments below you'll find: 
- full walkthrough demo 
- end-to-end guide
- my exact prompts and skills used 

Interested to hear, what other workflows would you like to see built with Claude Cowork? I'll be releasing more soon.

Big news ‼️ MCP Apps just landed in Cursor

Your Agent in Cursor can now render full interactive UIs mid-conversation

First it was ChatGPT Apps.. now its Claude, ChatGPT, VS Code, Goose, Windsurf, Cursor and many more.

Not only is this is a big deal for users as they no longer have to bounce between tools and can do everything directly from the AI interface

But its a big deal for companies  as well.

If you're Amplitude, you no longer need users to leave their workflow to check analytics. Your dashboard shows up inside Cursor, Claude, ChatGPT, VS Code - wherever your customer already works.

If you're Figma, your designs render inline while a developer is coding. No context switch. No "let me go check the mockup."

Really any company that has it's users using AI then having to switch to their platform - you now have a way to meet them where they already are.


As all of these are built using MCP Apps, it means you can build once and deploy across all the major AI platforms.

An exciting new distribution channel that didn't exist 12 months ago.


AI is becoming the operating system and companies are positioning themselves now be the tool that people choose to use.

I ran an AI acceleration session with the Noetic team in London last week.

Watching people transform their workflows in real time is one of the best parts of the job.


We last ran a session together in December and the amount that's changed since then is a reminder of just how fast this space is moving.

This time we covered what's cutting edge right now and most relevant to their business: 
Claude Skills, Claude Cowork, Gemini Deep Research, Image & Video Models, Twelvelabs, the Intent-Based Web with the latest developments with ChatGPT Apps & MCP Apps and even Openclaw.


But the aim of these sessions isn’t just to give an update on the latest tech, it’s for everyone to actually build with AI. 

What’s so good about these focused full day workshops is the amount of progress you can make when everyone is focused on getting practical and using the tools.

Everyone building Claude skills, testing image tools, practically learning how to use these agents. 

It’s great to see how people can, in just a few hours, transform their workflows and think of ideas how they can instantly automate their work and focus on higher value tasks.


Always great fun with Roslyn Casey Shaw and the Noetic team! Curious, engaged, very switched on and ready to try new things. It’s a pleasure to work with companies that are so forward thinking and ahead of the curve.

Claude Cowork is by far the most powerful productivity tool available right now for non-technical people

And most people still haven't heard of it

Here's what it is, how it works, and how you can use it to automate entire functions of your job


What is Claude Cowork?

Think of it as Claude Code (Anthropic's elite coding agent) but rebuilt for everyone who doesn't write code

The reason thats important to note is that the coding agents are by far the best at executing tasks, but for many people who aren't developers, they're complex to use 

So, Anthropic wrapped claude code in a simple desktop interface and called it Cowork.

It plans, executes (deciding all the tools it's best to use inbetween) and delivers finished work back to your files


These are the 4 tools it uses: 

... quick point: you want to really understand these as the better you can get a grip of the tools it has available, the better you can think of what use cases you can get it to do to eliminate whole functions of work

1. Skills 
- Reusable task instructions that teach Claude how to handle specific outputs - here are some examples of mine: newsletter writer, case study creator, sprint planner, higgsfield prompt creator, youtube description creator  
- skills allow you to consistently do tasks well. Build your own for anything you do repeatably 
- importantly, skills can be text describing a workflow but also files that give claude the right context any time the skill needs to be used

Connectors (MCP & MCP Apps) 
- These link Claude to your existing tools
- Google Drive, Gmail, Notion, Apollo, Clay, DocuSign, n8n etc. 
- Right now, you can see on appdiscoverability .com/track, 153 of the worlds leading companies have launched connectors to allow you to use their capabilities in Claude
- In Cowork, they gain filesystem access. Real integration, not just reading.

Sub-agents 
- For complex tasks, Cowork spins up parallel sub-agents running workstreams simultaneously. One analyzes your CRM while another drafts the email sequence

→ Commands 
- Saved workflows triggered with a /. Type /call-prep before a meeting, /write-spec for a new feature, /competitor-research for a new account. One word launches a full multi-step process. You build them once, run them forever.

Plugins 
- Pre-packaged bundles that wire up the right skills + connectors for a specific job function in one click. Sales, HR, legal, finance, ops - all ready to go.


The use cases here are endless

You can use Apollo + Clay to automate entire lead gen flows

Granola, G-drive & Gamma to review documents & call transcripts then create entire proposals 


I'm going to be creating demos of this some workflows this week

Shout if there are particular ones you want me to cover 

—
If you want to get daily or weekly updates on the Claude Connectors & ChatGPT Apps as they go live, sign up at appdiscoverability .com/track

JUST IN 🚨 Microsoft just dropped the biggest Copilot update since launch

And 99% of people missed the biggest news in the announcement:  

They launched 3 major announcements

1. Copilot Cowork

An AI agent that brings long-running, multi-step work into Microsoft 365 Copilot. Moving beyond prompts and responses toward execution that unfolds over time.

2. Copilot across M365 apps

Copilot now works alongside you in Word, Excel, PowerPoint, and Outlook - creating, editing, and refining content from start to finish inside your apps.

3. Agents in Chat - ‼️ with MCP App support ‼️ 

This is the one most people will missed.

Microsoft said agents in Copilot now support "open standards like Apps SDK and MCP Apps" - meaning your apps can "surface directly within chat, enabling live, interactive experiences where work actually happens."


If you're building an MCP App, this is huge.


400 million enterprise users just became the addressable market for every MCP app.

All knowledge workers inside every enterprise running Microsoft 365 is now a potential customer. 

Your app can now live inside the same tool people use to read email, join meetings, and manage their entire workday.

Adobe, Monday .com, and Figma are mentioned directly in the announcement. 

So this means when you build an MCP app, it can now be accessible across OpenAI, Anthropic, Cursor, VS Code, Goose and now Microsoft 365.

One protocol. Every major platform. Hundreds of millions of users



The companies building MCP apps right now aren't just shipping integrations. They're getting distribution through the biggest software ecosystems in the world.

What a time to build.

---
If you want to stay ahead of MCP Apps and get what apps are launching across ChatGPT, Claude, Cursor and now Microsoft Copilot straight to your inbox, sign up at appdiscoverability .com/track (link in the comments)

JUST IN 🚨 OpenAI drops plan for direct checkout inside ChatGPT. This is a huge deal for the future of ecommerce.
Where will check out happen now? Inside ChatGPT APPS!

Here's what's actually happening under the hood:

ChatGPT shopping worked like Google Shopping - users browse, compare, research. But almost nobody checked out. Only about 12 Shopify merchants ever went live with direct checkout. Out of millions.

But, there have been two problems:

1. Keeping inventory, pricing, and shipping synced via a product feed across millions of retailers is nearly impossible in real-time. Even Google gave up on this and just redirects to the retailer's site.
2. People prefer to buy where they already have accounts, saved cards, and loyalty programs. ChatGPT isn't that place, yet.

So OpenAI made the smart move. Shopping search stays - but it now redirects to the retailer's website to complete the purchase. Just like Google Shopping.

But here's the key part: In-chat checkout isn't dead. It's moving to ChatGPT Apps.

"Instant Checkout is moving to Apps, where purchases can happen more seamlessly." - OpenAI

Apps like Instacart and Target already handle full purchases inside ChatGPT. The app owner maintains inventory, pricing, and availability - so ChatGPT doesn't have to. That's why checkout works inside apps but couldn't scale across raw product feeds.

What we don't know yet is exactly how shopping search and apps will connect. Will ChatGPT automatically suggest a retailer's app when it finds their product? Or will search and apps remain separate experiences? OpenAI hasn't said.

But the direction is clear: if you want to transact inside ChatGPT, you need an app.


Now here's where it gets really interesting.

GPT-5.4 launched YESTERDAY with a feature called Tool Search.

Previously, ChatGPT loaded every tool definition into the prompt upfront. Expensive, slow, doesn't scale.

Now? The model gets a lightweight index of all available tools and searches for the right one on demand - by reading tool descriptions and metadata.

In testing across 36 MCP servers and 250 tasks: 47% fewer tokens, identical accuracy.

What that means in plain English: your app's description and metadata now directly determine whether GPT-5.4 even looks at your app when a user asks a question.


So for any company thinking about AI commerce and being present in ChatGPT & LLMs, the new playbook is becoming clear:

1. Submit your product catalog via ACP (Agentic Commerce Protocol) - so ChatGPT search can surface your products

2. Build a ChatGPT App - so customers can complete the purchase without leaving the chat

3. Optimize your tool metadata - because Tool Search decides whether your app gets found based on how well your descriptions match what people are asking for


---
If you want to track whats happening in the ChatGPT App Store daily, with all the new apps being released, you can do it at: appdiscoverability .com/track

now live! - ChatGPT App Store Discoverability Score 

you can now get the score for every live ChatGPT App

- click on any app
- select 'Visibility Score' 
- see how its performing
- see recommendations to improve it

right now, the ChatGPT App Store is the only place your app can be discovered so its critical to ensure you're as optimised as you can be for discoverability.

check it out here: track .appdiscoverability .com

Wow, this is insane... we built an agentic product pipeline with OpenClaw & Remotion that ships the code and records a DEMO VIDEO of the finished feature - you one click approve & the feature is live. Here's how it works:

In this case, we were adding cross-platform intelligence to appdiscoverability .com/track - letting you see which apps are on both ChatGPT and Claude. A whole new feature pushed live through this flow.

The video that Openclaw created with Remotion for us to review (a platform that creates video with code), is what you're watching below...


I couldn't believe this worked when we tested it to be honest. This is what happens when you push Openclaw & Remotion to the edge

This is exactly like having a developer reporting to you who does the work and then gives you a personal demo before you approve


The reason we built this is that reviewing AI work is a pain. You'd prompt something, but then spend ages checking different formats trying to figure out what actually changed.

Now it creates a presentation of the updates - video walkthroughs, or screenshots for quick changes - so you can clearly see all the work that's been done.


Here's how the pipeline actually works:

1. Product spec - the agent writes the full PRD with user stories, acceptance criteria, and estimated hours. Posts it in chat and waits for thumbs up before any code gets touched

2. Code - OpenClaw spawns a Codex sub-agent to write the code. Cheap model does most of the work, Opus steps in when something gets tricky

3. PR + CI - ships as a GitHub PR with linting, type checks, and build all automated

4. Vercel preview - every PR auto-deploys a preview URL so you can click and see it live

5. Demo video - this is where it gets cool. The agent opens the preview in Playwright, walks through every user journey, and then uses Remotion to actually code a video demo of the walkthrough. For smaller updates, it can also just capture screenshots with Playwright if that's all you need

6. Review & merge - you watch the demo and approve to make updates or directly push the change live


Honestly this is probably what building software looks like for everyone in 18 months 

What a time to be alive

Full detailed guide in the comments below on how we did it 👇

Just 3 weeks ago, I sat down with Nebius's Co-founder Roman Chernin to break down their $3 billion Meta deal.
And just yesterday, they announced a new 27 billion dollar deal... 

$12B in dedicated capacity across multiple locations - one of the first large-scale deployments of NVIDIA's Vera Rubin platform, starting early 2027.

Plus Meta has committed to purchase up to $15B more in additional compute across upcoming Nebius clusters over 5 years.


However, what a lot of people miss about Nebius is that these hyperscaler deals aren't their only business function. 

In fact, every dollar from Meta and Microsoft buying their GPU capacity, funds what Nebius' core value position - a full-stack AI cloud for the rest of the market.

Nebius' full offering:

-> Bare metal infrastructure for hyperscalers
-> Multi-tenant cloud for AI startups and digital-native companies
-> Managed inference and post-training (Token Factory) for enterprises building AI products and agents

In our conversation, Roman broke down how this stack serves each layer of the market - and why any company running closed source models today will eventually need to shift to open source at scale to optimise cost and while maintaining quality. 


Roman's prediction: "If you think enterprise adoption will take a long time... you'll be surprised how fast they move."

$27 billion in new infrastructure doesn't get built without a confident bet there is somewhere for that compute to go.


Full conversation below 👇

We had some exciting news that Vincent McLeese and I will be giving two talks at MCP Developers Summit in New York on April 2nd! 


1. 'If the LLM Can't Find You, You Don't Exist'

We've been building apps and working with companies to launch them since the MCP App standard was released.

And we always get asked the same question:

"Once we've launched, how do you actually get our app discovered?"

We've spent the last 6 months researching, collecting data and developing tools to solve the ChatGPT App & MCP App discoverability problem.

We break down everything we know about discovery in the app store and organically.



2. 'Building ChatGPT Apps: Principles for a New Kind of Interface'

ChatGPT apps aren't websites or mobile apps.

They're lightweight tools that live inside a conversation, invoked by a model. You don't own the screen. The conversation defines the flow.

If you approach them with the wrong mental model, you'll build the wrong thing.

We'll share the design principles and lessons we've learned - from scoping what belongs in a conversation to making sure users actually come back.



At Ghost Team, we've been building ChatGPT apps and MCP apps for enterprises across industries.

We're bringing everything we've learned into these talks.

One of the tools we've built along the way is the first ever MCP App Tracker - this is the largest dataset available on ChatGPT App and Claude Connectors.

It gives a unique view into what's actually working, how discoverability is shaping up and keeps you up to date with what's happening. Check it out here: https://lnkd.in/ejnAZGD3



We're very excited about the direction for MCP Apps and even more excited to be sharing what we've learned in April.

If you're in New York, don't miss it!

Let me know if you're there - would love to grab a coffee ☕️

"My idea is too niche for a ChatGPT app."

Meanwhile, a Korean smart farm just launched a ChatGPT App helping people order fresh lettuce

ChatGPT has +850 million weekly users. If there's demand for pesticide-free salad greens, there's demand for whatever you're building


Jokes aside, we're at the beginning of seeing a long tail of AI apps just like we've seen in mobile app stores

The more niche the better for capturing target intents

taplio

The Battle

Anna Erlenkamp

Elliot G.

Followers

Posts

Engagements

Likes

Performance ScoreElliot G.'s LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Anna Erlenkamp

Elliot G.

Followers

Posts

Engagements

Likes

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

Ready to grow your LinkedIn presence?

Performance ScoreElliot G.'s LinkedIn Performance