Federico Martelli

I'm Federico we raised €5M in 36 hours, the fastest pre-seed in Europe and joined Forbes 30 Under 30 Europe 2026 shortly after I'm starting on X to share what I've been building and what I've learned along the way but first, a bit about me 27, Italian, CEO of Forgis, based in Zurich 4 months after founding, we had 5 term sheets from international funds won the ABB Startup Challenge against hundreds of startups and got deployed inside Fortune 500 plants all while obsessing over one question: why is the West handing its industrial base to China, and what can actually be done about it I've been writing about this on LinkedIn for the past year 10,000 followers, 800K monthly impressions turns out a lot of people in manufacturing, automation and deep tech care about this stuff I'm here to build the same thing on X expect real insights on manufacturing, industrial AI, robotics, reshoring and the geopolitics of production no fluff, no startup motivation content just first principles thinking from someone actually building in this space if that's your world, follow along

@bellosoooo early stage means whatever it takes. the range is part of the job description.

The life of a bootstrapped founder. Morning: closing a partnership deal Afternoon: cleaning boat for quick cash Gotta love the game

@egbennis compounded curiosity is the right framing. talent is just the name we give to invisible reps over time.

the people I've watched actually win at building all share one boring trait. they go further on every subject than anyone around them needs them to. not perfectionist-paralysis. more like curious-relentless: reading the second paper, running the extra test, rewriting the email twice, pulling the thread when everyone else has moved on. at month 3 you can't see the gap; by year 5 it's unbridgeable, and people call it talent because we don't have a word yet for compounded curiosity.

@jenngrannen deployment as the training loop is the right architecture. production data beats synthetic data every time.

Very excited about this work on data flywheels for improving VLAs across a variety of tasks. Deployment is not the finish line -- it's the unlock to a whole new source of data!

@astnkennedy delegating execution is fine. delegating the thinking is where the sharpness starts to erode

I'm 22 years old and Claude Code is deteriorating my brain. Every single day for the last 6 months I've had 6 to 8 Claude Code terminals open, waiting for a response just so I can hit 'enter' 75% of the time. And it's doing something to me. In convos with a couple of friends, it's been a point that's been brought up pretty frequently. None of us feel as sharp as we used to. I don't know if it's just us, or others in their 20s are feeling the same thing, but it's something I've been thinking about a lot. P.S. I know this is a problem with my reliability/usage of it, not Claude Code itself, but the effects are real nonetheless

@_asadmemon best hardware testing rigs are always the ones held together by ingenuity and zip ties

You all seem to like drones A LOT!! I need to test Mighty Camera on drones with vibrations etc. This poor whoop 😅

@asimahmed @NianticSpatial @SpatialJP world models are the prerequisite for physical AI that actually works. spatial understanding before autonomy.

at @nianticspatial, we're building the real-world model for physical AI. we believe the next generation of AI will move beyond the screen, into real-world environments where some of the hardest problems need to be solved. with high-fidelity gaussian splats as the foundation – robots, AI agents, & autonomous systems can see, localize, understand, & operate in the environments that matter most.

@MorganVonDruitt tool sprawl is a proxy for unclear ownership. the audit reveals the org chart problems, not just the spend.

Stack consolidation is the most underrated cost-cutting move at seed-stage SaaS. Here is the playbook. Step 1: pull the founder into a Zoom. Open every SaaS login one by one. Note which tools the founder doesn't recognize. (You'll find 3-5 vestigial subscriptions.) Step 2: ask each function lead 'name three tools you'd cut and three you'd keep.' Anything in the 'cut' column for two leads is a candidate for kill. Step 3: identify duplicates. Most teams pay for both Zapier AND Make. Both Loom AND Vidyard. Both Notion AND Confluence. Pick one per category. Step 4: cancel the kills. Migrate the data within 7 days. Don't dither. Typical recovered budget: 20-40% of total software spend. For a 15-person seed-stage SaaS; that's $30-80K/year. Reallocate to AI seats and content.

@SGhashghaei robotic depopulation at enterprise scale is exactly the kind of unglamorous work that compounds into a real moat

When one of the largest companies on the planet asks if you can depopulate network boards and server boards. Uhhhh absolutely. Tuurny.com

@daniel_dhawan @speedrun the reps before the breakthrough are not wasted. they are the only reason the breakthrough holds.

My first 6 years as a startup founder: - Failed with 4+ startups - Ran out of money multiple times & had $15K credit card debt - Was rejected by Y Combinator 8 times - Got 200+ rejections from investors My last year as a startup founder: - Got into a16z @speedrun & raised $18M - Moved to SF & got O1 thanks to @tmhammer & @lighthousehq_ - Scaled Rork to #1 AI mobile app builder in the world The average journey to a $1B company takes 10 years. I’m on year 7. Keep building.

@jakeottiger the best technology adoption stories are never about the technology. they're about the operator.

daniel is such a cool dude. i cannot wait to share his story with all of you soon. he runs a cleaning supply distributor and a vintage clothing store in LA. he was also a professional soccer player. we are rolling out Codex to him to help him run his businesses better, reduce monotony, and grow with less stress. in these documentaries, there will be no mention of “ai” nor “ai transformation” just incredible people and businesses that deserve to be heard.

@saadixsd AI as decoration gets cut in the next budget cycle. AI as the engine compounds every quarter.

Most founders treat AI as a decorative layer. Here is why that fails. Most founders are stuck asking how to add AI to their product. I've seen it first-hand this tactical error that leads to bloated, purposeless features. Instead of looking for a place to insert a chatbot, you should be designing high-leverage workflows. AI is most effective when it is the engine of a specific, repeatable process rather than a standalone gimmick. The shift from feature-thinking to workflow-thinking requires a structural approach. First, map every manual step in your core operation. Second, isolate the specific task where human throughput is the bottleneck. Third, architect an automated sequence where AI handles the data transformation between those steps. Focus on systemic efficiency over trend adoption. Build the infrastructure that allows your business to scale without linear headcount growth. When you solve for the workflow, the AI integration becomes obvious.

@tizimmer copying the motion without the platform underneath is just expensive professional services with better branding

forward-deployed engineer roles are up 800% this year everyone copying the Palantir playbook of embedding engineers with customers but here's the thing: Palantir built opinionated platforms first. the engineers deploy those. most companies skipped that step. they're just doing custom dev work and hoping patterns emerge later.

@harshitduggal5 automating a broken process doesn't fix it. it just scales the damage faster and more efficiently.

You do not have a “distribution problem.” You have a precision problem. Bad ICP. Bad account list. Bad trigger timing. Bad follow-up. Bad conversion from positive reply to booked call. That is where revenue dies. Most teams spray more outbound because it feels like action. It is not. It just creates more noise from the wrong people. You do not need a bigger list. You need one narrow ICP, a cleaner buying trigger, tighter follow-up, and a system that turns interest into booked pipeline before it goes cold. Your AI SDR will not save a broken GTM motion. It will just automate the leak faster.

@DomJoLuna deployment without governance is just technical debt at enterprise scale. visibility is the real product gap.

Gartner says 40% of enterprise apps will embed AI agents by December. Up from 5% last year. That's an 8x growth rate in 12 months with almost zero governance infrastructure in place. makes me realize the industry doesn't have an AI problem. They have a visibility problem called AI.

@richardzphotoz conversion and awareness compound differently. mixing the metrics is where most strategies break down

There’s usually two types of marketing. One that converts into new customers and one that’s for brand awareness. Both equally important.

@InfiniteJulian0 memory at the edge is the real unlock. intelligence that travels with you changes the whole stack.

visited the Mira HQ yesterday, super cool product & bullish on the team! some of the top use cases: - memory based on transcriptions (builds a second brain for you) - live translation - view notes when giving a talk/speech - deploying agents at any moment

@stash_pomichter persistent spatial memory is the missing layer. context windows were never built for production robots.

Announcing a new Memory system for robots on Dimensional Robots in production generate thousands of hours of video, lidar, odometry, far too large to fit into your Agent context SpatialMemory2 builds a multimodal data store in latent space for your Agents Fully open source

@dariusparzygnat AI compresses the cost of iteration. the constraint was never capital, it was always speed of learning.

With AI, software development only feels expensive if you think your time is worthless

@VectorWang2 zero-shot sim-to-real without fine-tuning is the unlock. domain randomization doing the heavy lifting.

Catching a flying ball is hard. What if with a flat plate? 🏓 Our work at RSS’26 shows it’s possible through Zero-Shot Sim2Real With Domain-Randomized Instance Set (DRIS), we catches different kinds of balls without any real-world fine-tuning 🔗 rice-robotpi-lab.github.io/DRIScatch/

@henrytdowling subagents are just division of labor applied to intelligence. same principles, different substrate.

Some heuristics for when to use subagents - you want to use a smaller model - you have too much context for the main agent to handle, rlm-style - there is a very natural and small interface for the task