Articles by Daniele
-
Mastering Prompt Performance: measure and enhance your AI assistant's efficiency and reliability
Mastering Prompt Performance: measure and enhance your AI assistant's efficiency and reliability
Creating a prompt is just the first step in building an effective assistant that works for the purpose you had in mind.…
1 Comment -
Advanced prompting techniques: Code Interpreter
Advanced prompting techniques: Code Interpreter
Ghost operates in an industry where turnarounds are pretty quick. Sometimes, listings can be gone in a matter of…
1 Comment -
Advanced prompting techniques: function calling
Advanced prompting techniques: function calling
tl;dr check out the Jupyter notebook! LLMs cannot rely on real-time knowledge. For example, an AI can't tell you what…
1 Comment
Activity
8K followers
-
Daniele Bernardi shared thisAI costs look cheap. Then usage starts. The story this week is companies moving from tokenmaxxing to tokenpocalypse. A lot of teams treated AI like a flat SaaS seat. It isn't. It's more like a taxi meter left running in traffic. - Token costs = the bill for words in and words out. - Inference costs = the compute bill every time the model thinks. - Orchestration costs = the extra plumbing to route, retry, log, and chain tools. - Total cost of ownership = all of it, plus humans checking the work after. This is why the cheap demo gets expensive fast. One prompt becomes five. One answer needs review. One agent loop quietly eats snacks all day and nobody sees the reciept until month end. AI is great, don't get me wrong. But the biggest cost is often not the model. It's the mess around the model. The same will happen in every market. The winners won't be the teams with the most AI. They'll be the teams with the cleanest cost per useful outcome.
-
Daniele Bernardi shared thisAI is getting cheaper. And more expensive at the same time. That a16z token efficiency chart is the part most people miss. • Token cost means what you pay per word chunk. Cheap per token looks great on a pricing page. • Inference cost means what it costs to get one answer. Like a taxi meter running while the driver takes the long way. • Orchestration cost means all the extra glue. Routing, memory, retries, evals, logs. The hidden line items on the reciept. A lot of AI teams are saving on the model and losing on the system around it. One agent writes 200 words. Behind it, 20k tokens burn in planning loops, tool calls, and self-checks. Great tech, don’t get me wrong. But total cost of ownership is the full supermarket bill, not the price of one snack. The same will happen everywhere. The winners won’t be the ones with the cheapest model. They’ll be the ones that get the most useful work from every dollar.
-
Daniele Bernardi shared thisAI got cheap. The bill didn’t. Chamath said his company’s AI token costs are doubling every 45 days while output is up around 5%. That tracks. - Token cost means the meter running each time the model reads and writes words. - Inference cost means paying every time you ask the model to think. - Orchestration cost means all the extra glue: retries, memory, tools, routing, logs. - Total cost of ownership is the full supermarket receipt, not just the item on sale. AI in prod feels a bit like kids eating snacks. One bag looked cheap. Then you see how fast the pantry empties. The weird part is the model is often the cheap bit. The real spend shows up in all the extra turns, longer context, agent loops, and humans checking the work becuase they still have to. Great tech, don’t get me wrong. Funny how AI is starting to look less like software you buy once, and more like a utility bill that learns how to grow.
-
Daniele Bernardi shared thisAI got cheaper. The bill didn’t. Palo Alto Networks CEO Nikesh Arora said AI token costs still need to fall 90% for broad business adoption. That tracks. - Token cost means what you pay per chunk of words. That part may drop. - Inference cost means what you pay every time the model does work. Like taxi rides that never stop. - Orchestration cost means all the extra plumbing around it. Routing, retries, evals, guardrails. - Total cost of ownership is the full supermarket receipt, not just the milk. Funny how cheaper models can still make AI more expensive. Because once it works, teams use it more. Then more again. Like kids finding the snack drawer. The hidden cost in AI is not the answer. It’s all the machinery needed to get one good answer, repeatably, at work scale. AI will spread fast. The winners may be the ones who make the bill feel boringg.
-
Daniele Bernardi shared thisAI pricing is getting exposed. Everyone is still selling tokens, credits, seats. The buyer wants one thing: the work gone. We saw this with low-code too. People said they wanted to build. Mostly they wanted the software to exist. The same will happen with AI workers. The winners won’t meter intelligence. They’ll sell finished work.
-
Daniele Bernardi posted thisVibe coding is real. And maybe that’s the problem. - It collapses the distance between idea and demo - It also collapses the distance between demo and delusion - We’ve seen this movie before with no-code: amazing for prototypes, weak for systems that need to survive contact with reality AI is great, don’t get me wrong. But the moat isn’t prompting faster. It’s structure, context, and the boring parts people skip. V1 is cheap now. V10 is where the company is.
-
Daniele Bernardi posted thisVibe coding is not the story. The story is what happens after. - anyone can ship a first draft now - very few can own the architecture - even fewer can maintain the mess they prompted into existence We’re watching software get cheaper to create and more expensive to understand. Funny how low-code died on maintenance too. The same will happen here. Build is abundant. Ownership is the moat.
-
Daniele Bernardi posted thisAI spend is starting to look familiar. - teams brag about token volume the way SaaS used to brag about seats - the bill grows faster than the outcome - everyone says costs will fall, then usage eats the savings We’re building the opposite at Toolhouse. Workers that do the job, on a schedule, with clear output. Funny how the AI race may not be about smarter models. It may be about escaping the new seat-license tax.
-
Daniele Bernardi shared thisI modelli stanno diventano troppo costosi. Secondo un sondaggio di UBS (sì, proprio loro), il 60% delle compagnie sta già passando a modelli cinesi e/o limitando la spesa AI delle compagnie. Noi ti permettiamo di controllare i costi. Le compagnie che usano Toolhouse risparmiano fino al 75% ogni mese. Se sei interessato commenta e ti spiego come.
-
Daniele Bernardi liked thisDaniele Bernardi liked thisEvery agency has a dream client category. For a lot of us in the performance marketing world, it's automotive. Today I'm excited to share that Truman Motors is officially an icepop client. Truman builds custom Super Duty trucks engineered from the frame up. Super single conversions. LiquidSpring suspension. 44-inch tires. These aren't modified trucks. They're reimagined ones. A brand this bold deserves creative and media buying to match. We're running their Meta ads and couldn't be more excited to dig in. Big things ahead. Welcome to the family Truman Motors. #MetaAds #Automotive #DigitalMarketing #icepop #PerformanceMarketing
-
Daniele Bernardi liked thisDaniele Bernardi liked this🚨 Don't Enter the USA alone 🚨 Most brands entering the US make the same mistake... They try to figure it out solo. No local partner with a deep understanding of the market. No playbook for US E-Commerce No connections into US retailers. What will your team salaries cost you to figure it out? A lot. This is where we come in. We've already solved the complexities of entering and scaling a brand in the USA. We start with entry. Our supply chain headquarters in Philly gets your product off the boat and into a real US warehouse, fast. Once you're in, the real fun begins. We launch your brand on Amazon and get you into Retail chains across the country. The time is now to enter the largest consumer market in the world. Just don't do it alone.
-
Daniele Bernardi liked thisDaniele Bernardi liked thisThere's never been a better time to post content as a founder. Which is why I am excited to finally publicly talk about a project we have been working on. Over the last 4 years, the big bet we've taken has been around how powerful personal branding is: We've created: 1. 100,000+ content pieces 2. Millions and millions of impressions 3. Millions of dollars in revenue online 4. 400 of the best founders and CEOs personal brands in the world And we just did something with everything we've learned. While building all of that, we were also building an AI ecosystem to run it that includes Agents for: -> Strategy -> Topics -> Writing -> Optimization These are live across dozens of founder accounts right now. And because it's watching all of those accounts at once, it catches anomalies and patterns that no single founder could ever see on their own. A few months ago we consolidated everything into one platform. We're calling it the Content Suite: Here's the entire system (bookmark this): 5 layers that feed each other: 1. Brand profile (it learns who you are and what you'll say) 2. Topics (it finds the ideas worth posting) 3. The interview (it pulls the thinking out of your head) 4. Structures (it matches your idea to the viral format performing right now) 5. Optimization (it tracks what actually worked) Then optimization loops back into topics with sharper data. That's the flywheel. Every post gets built the same way: It pulls the idea out of your head, matches it to the viral content structure performing right now for that exact type of idea, then writes the post with you, piece by piece, in your voice. Most founders skip straight to writing, which is the biggest mistake you can make. Sending a raw idea to ChatGPT skips the interview and the structure, and you end up with content that sounds like everyone else's. This is some of the work I'm most proud of, and we're only getting started. New modules are in the works right now. I'll be talking about this a lot more on this account, and I'll be quoting this post with updates as we build. Below you can see how it works + a content piece that went live on a nearly unused account. Excited to share more :)
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
Licenses & Certifications
Publications
-
iPhone X surpasses Samsung and Google on repeat purchases
VentureBeat
See publicationOur team took a look under the hood to see how iPhone X consumers are engaging with retailers compared with buyers wielding Samsung’s Galaxy S8/S8+ and Google’s Pixel 2. While these flagship devices account for just 1-3% of overall installed base for the retailers we examined; the differences in how these consumers engage are significant.
Courses
-
Big Nerd Ranch Android Engineering Course
-
-
Big Nerd Ranch iOS Engineering Course
-
-
Media Training
-
-
Public Speaker Training
-
Projects
-
Mobile Personalization Solutions Launch
I'm part of the cross functional team that launched our Mobile Personalization Solutions, part of our broader Omnichannel initiative.
- Help Marketing identify trends and deliver consumer analysis
- Work with Sales to assist them during pre sale
- Write technical documentation and client-facing collaterals, including help and training material
- Present at webinars
- Conducting User Acceptance Tests
- Ensure our partners comply with our Client Integrations…I'm part of the cross functional team that launched our Mobile Personalization Solutions, part of our broader Omnichannel initiative.
- Help Marketing identify trends and deliver consumer analysis
- Work with Sales to assist them during pre sale
- Write technical documentation and client-facing collaterals, including help and training material
- Present at webinars
- Conducting User Acceptance Tests
- Ensure our partners comply with our Client Integrations requirements
- Prepare prototypes and proof of concepts for cross-platform personalization -
Subject Matter Expert, Mobile
- Work on our beta SDKs for iOS and Android to match customer requirements
- Partner with Product Managers to identify product enhancement and to influence roadmap
- Drive User Acceptance Tests
- Gather and document best practices
- Early adopt any new feature before public launch
- Partner with Client Success to identify and resolve known bugs
Honors & Awards
-
Winner - OpenAxel Startup Innovation Competition
OpenAxel
Awarded as one of the most innovative startups of Southern Europe
Languages
-
English
Full professional proficiency
-
Spanish
Professional working proficiency
-
Italian
Native or bilingual proficiency
Recommendations received
-
LinkedIn User
1 person has recommended Daniele
Join now to viewOther similar profiles
Explore more posts
-
Ioana Dayagi Stamate
SF Tech Week kicked off yesterday, and I hopped over to a Reforge session on prototyping. I wanted to soak up the SF energy and see what’s new. Sachin Rekhi shared highlights from his new Reforge course, plus Brian Balfour announced Reforge Build’s tools. It is clear that Reforge are expanding from being a teaching only shop to building the tools that can help product teams. Here are my takeaways from the talk and from my experience with prototyping (yes, I needed to join this talk to realize it's worth sharing my learnings): 🐒 - Pair prototypes with analytics to get earlier user validation. Watching what people do before you ship cuts waste and validates faster. You can import your product style and layer new flows right on top. This was a new insight for me that makes complete sense—moving the validation phase up. 🎁 - Use the AI vast examples. I avoid providing overly detailed specs when I prototype because I want additional opinions during the building process rather than being stuck with one approach. Many tools now offer options like variations (in the new Reforge Build program), inspiration on Magic Patterns or just ask for several options. I loved how Lovable presented several options with pros and cons, letting me decide what I preferred. 👭 - Prototyping as a team is still early, but it works if you design for it. Also, handing off a prototype to engineering might not be fully internalized everywhere, but I believe as prototypes start to replace parts of PRDs, this will become more standardized. Q&A is always my favorite part. Hearing other people’s constraints sharpens my thinking. 😍 How I build from scratch 😍 : • I start with the main screen where the core activity of your app happens. Connecting with AI tools and databases is essential to identify the small details you'll need to point out or consider. Inquire about preferred design patterns to create a better app. • Then I add everything else along the way—navigation, home page, save functions. • After navigation changes, test all paths, as AI tools aren't very good at maintaining consistency. 𝐒𝐮𝐦𝐦𝐚𝐫𝐲 We’re not at “prototype → deploy” yet, but PMs are much more empowered to test paths, pressure‑check prioritization, and change our minds fast. Go Build!! #Prototyping #ProductManagement #SFTechWeek #Reforge
-
Felicia Curcuru
I recently sat down with the SF Business Times to talk about how Binti is bringing AI to social work. We discussed our partnership with Anthropic and the incredible responses we’ve already heard from social workers who are using Binti AI. We also talked candidly about data privacy, ethical considerations, and how we’re approaching AI in government. And of course, we talked about the Bay Area — whether it’s still the tech hub and the best place to build a company (spoiler alert: I said yes). Read the full piece below! ✨ #AIforGood #EthicalAI #GovTech #BayArea #Anthropic #Binti
4 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content