Filippo Pietrantonio (@FiloPietra) header
Filippo Pietrantonio (@FiloPietra) profile picture

Founder https://nitter.gq/t.co/nJoPMdwLwr | Building & deploying aiOS for organizations

Joined October 2011
Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)Photo by Filippo Pietrantonio (@FiloPietra)
Most systems are outdated and built for the pre-AI era. Companies who are AI-first are seeing massive returns and gains in productivity, security, reliability, scale, cost reduction, time saved, and much more. My thoughts:
1
1
36
What I've been saying for a while now. Human-written code is inefficient and insecure in an AI world. You cannot afford to have outdated systems.
All, I repeat ALL, existing software needs to be rewritten. Because this software was written by humans it is both inefficient and insecure. The cost of inefficiency and insecurity in a pre-AI world was moderate, but in a post-AI world it will destroy your company. This new effort should not be a bottoms up exercise where engineers get to have an opinion about how they set up their repo or which coding assistant to use. This needs to be done in a new, top down system of record that focuses on auditability, drift detection and verification among other things. Every company needs to do this asap as today has shown. If you want help: 8090.ai
21
Highest paid jobs in 2027-2028: - forward deployed engineer - marketing engineer - IRL event manager - tech content creators
40
AI adoption seems slower inside PE-backed companies than in other businesses. But when a portfolio company builds AI into how its teams work, the upside can be remarkable. Finance can turn work around faster. Deal teams can follow up with more buyers while interest is still fresh. Managers can see where work is stuck before the next weekly meeting. At Mesh Flow, I start with the work people already do every week. For one brokerage, that meant building AI into how the valuation team prepares financials, flags missing information and drafts prospectuses. The team reviews the work and approves updates in the CRM it already uses. That’s how an improvement in one workflow can open up capacity elsewhere. If the valuation team spends less time chasing missing information, it can move more deals through preparation. And when the status is visible in the CRM, managers can step in earlier when a deal stalls. If I were reviewing AI progress with a portfolio company this quarter, I’d ask two questions. Which workflow does your team run differently today? What number moved because of it?
25
Great sign. Companies are really doubling down to becoming ai-native
US investment in IT equipment and software hit a record 5.35% of GDP last quarter. At the peak of the dot-com bubble, it was 4.46%.
42
the need for data is only gonna increase in the next years. in the age of AI, data is one of the most crucial thing to have. The more quality data you have about your business, operations, workflows, and processes, the easier you can train models, automate with AI, and have clear evals.
24
Crazy!!!
BREAKING: Nvidia, $NVDA, CEO Jensen Huang has resigned as CEO, President, and member of the Board of Directors, effective immediately.
37
Businesses that treat agents as first-class customers have a shot at a wave of growth. Those that don’t must now fight for bots’ business instead of keeping them out.
Move over, B2B and B2C - the B2A economy is here Agents are the primary users of the Internet, and soon will be the primary customers for every product Our thesis @a16z on the massive opportunity here 👇 (cc @venturetwins)
35
Quality of data will be more and more important. f you're a company sitting on years of proprietary operational data, you have two options: 1) use that data to post-train an existing AI model for tasks unique to your business. Think of @harvey, but built internally around your own processes. Challenge: you need engineering talent, infrastructure, and ongoing maintenance. 2) license that data (dm me) to AI labs and get paid for helping frontier models become better at real-world tasks. You don't need to build or host anything yourself, but you're betting on external models improving rather than owning the technology.
The Machine Learning market is seeing accelerating growth. The global machine learning market is now projected to grow from ~$100 billion in 2025 to ~$684 billion by 2033. This implies a CAGR of +27.2% from 2025 to 2033. As AI evolves, the focus is shifting toward AI models that can continuously learn, retain knowledge, and adapt to new tasks. This represents a fundamental shift from AI systems that rely primarily on knowledge acquired during training to systems capable of improving through experience. As adoption accelerates, demand for machine learning software, hardware, and services is expected to increase substantially. We are watching this market.
26
Is there any cloud service that has PS5 turned on 24/7 for people to play games from their PC at any time? Similar to hosting your own GPUs vs renting them on the cloud. Would be massive @levelsio
Set up PS5 Remote Play on MacOS so now I can play GTA VI when it comes out on my MacBook Pro even when traveling! You do need to bring your Sony controller because it won't accept any other regular controller
1
69
Great way to hedge your risk as a small business. And you don't need a financial institution to create that product for you anymore.
Mammoth Rock ‘n’ Bowl is a restaurant and bowling alley in Mammoth, California, one of the top ski towns in America. Like the hotels, rentals, and shops around it, the business runs on winter tourism. A bad snow year hits revenue hard. Rock ‘n’ Bowl is especially exposed since they also sponsor a youth snowboarding team. So rather than wait and see, co-owner Ashley Marks asked Kalshi for a market on Mammoth Mountain December snowfall. We built it, and she’s trading $5,000 for a $31,250 payout if there’s weak snow. "We created the snow market with Kalshi because like other mom and pop businesses, we live with seasonal snow risk and wanted to give the community a way to offset some of that uncertainty." - Ashley If the snow comes, business runs like any other winter, and she loses the $5,000 trade. If it doesn’t, the Kalshi position pays out, giving Ashley $26,250 to help offset lost tourist revenue. Either way, Ashley's business is protected.
41
Anthropic's employees will cash out an average of $41M EACH out of the IPO. That is crazy
BREAKING: 1 in 70 Americans worth over $100 million could be Anthropic employees after IPO Nearly 100% chance Anthropic officially announces an IPO this year
49
Filippo Pietrantonio retweeted
Some observations from the claude fable + mythos 5.2 launch 1) fable scored 41% on a benchmark of real freelance tasks. fable 5.1 scored 18%. This makes real, process-free work easier to automate. 2) an uber engineer gave fable a job that ran for 10–12 hours. This means a small business owner can give a complex list of tasks to work on overnight and check it off in the morning. 3) claude writes over 90% of the changes anthropic merges into its core codebase, with humans guiding it. We'll see more and more companies with 4 engineers doing the work of 40. 4) fable coordinated 1,000+ subagents for 8 days and built a prototype operating system that boots on a laptop. Imagine re-working entire outdated IT systems with a few all-nighters by Fable, without hiring extra engineers. 5) fable designed a voice hub, a speakerphone and a handheld virtual pet from one brief. All three worked without electrical changes. Fable allows to develop better hardware and prototypes, especially when starting out. 6) one tester said fable realized its estimate was wrong and held off giving a new number until it had measured the change. If you’re building ai into finance or legal work, that’s the kind of behavior you want to see on an ordinary tuesday. 7) 1,000 agents turned the proof of fermat’s last theorem into something a computer could check in 30 hours. Sounds serious. 8) fable and mythos are the same underlying model, with different safeguards and access rules. This is important because you can use the same intelligence but applied differently. We'll probably see more models like this. 9) both models more often said they suspected they were being tested than anthropic’s other recent models.... 10) on one coding benchmark, fable matched opus 5.5’s best score at roughly half the cost per task. i think the next thing people will get good at is knowing when to spend $5 on a hard problem and when a 5-cent model will do. Let's see OpenAI's response...
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
5
1
6
502
What are the most AI-advanced companies doing differently? I broke down a recent BCG research into a practical guide to picking your first workflow, measuring what improves, and deciding what AI can do on its own.
2
1
2
168
Filippo Pietrantonio retweeted
Only in Dubai can you work from a coworking space that feels like Bali, go to Chinatown for lunch, grab a 1 AED karak, then get stuck in traffic behind a Lambo, then close a deal over a $3 shawarma, drive past an indoor ski slope, and then play padel with a 21yr old millionaire
31
30
5
520
36,263
Some observations from the claude fable + mythos 5.2 launch 1) fable scored 41% on a benchmark of real freelance tasks. fable 5.1 scored 18%. This makes real, process-free work easier to automate. 2) an uber engineer gave fable a job that ran for 10–12 hours. This means a small business owner can give a complex list of tasks to work on overnight and check it off in the morning. 3) claude writes over 90% of the changes anthropic merges into its core codebase, with humans guiding it. We'll see more and more companies with 4 engineers doing the work of 40. 4) fable coordinated 1,000+ subagents for 8 days and built a prototype operating system that boots on a laptop. Imagine re-working entire outdated IT systems with a few all-nighters by Fable, without hiring extra engineers. 5) fable designed a voice hub, a speakerphone and a handheld virtual pet from one brief. All three worked without electrical changes. Fable allows to develop better hardware and prototypes, especially when starting out. 6) one tester said fable realized its estimate was wrong and held off giving a new number until it had measured the change. If you’re building ai into finance or legal work, that’s the kind of behavior you want to see on an ordinary tuesday. 7) 1,000 agents turned the proof of fermat’s last theorem into something a computer could check in 30 hours. Sounds serious. 8) fable and mythos are the same underlying model, with different safeguards and access rules. This is important because you can use the same intelligence but applied differently. We'll probably see more models like this. 9) both models more often said they suspected they were being tested than anthropic’s other recent models.... 10) on one coding benchmark, fable matched opus 5.5’s best score at roughly half the cost per task. i think the next thing people will get good at is knowing when to spend $5 on a hard problem and when a 5-cent model will do. Let's see OpenAI's response...
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges. They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
5
1
6
502
I've been experimenting with Devin Max for the past week after I was given a free month and I must say it's impressive. It keeps great context with the "wiki" mode, works extremely well in the cloud, follows best practices & vulnerabilities checks all the time, and much more. I can one shot issues or new features and it opens many PRs, merges them, and makes sure everything is stable before pushing to production (it runs it's own test from the terminal + its computer, which is great!). I can do the same things with Cursor, but it just takes more time and manual checks... Will probably keep both but Devin is a great addition to the stack! @cognition @devindevelopers
57
Am I the only one not getting why you would use @WisprFlow? Like I don't want everyone around me listen to what I wanna do... and also it would be extremely disturbing to other too. Am I missing something?
2
67
Why bother with AI? Just buy donuts and go about your day
BREAKING: Krispy Kreme ($⁠DNUT) stock soars as high as 42.9% in today's trading session.
47