Webrobot Ltd

04/08/2026

Tutti ripetono la stessa cosa: l'AI è la rivincita degli introversi.

È vero a metà. Ed è la metà sbagliata quella interessante.

L'AI non ha spostato il vantaggio dagli estroversi agli introversi. Ha riprezzato le funzioni cognitive sottostanti — e il riprezzamento taglia trasversalmente entrambi i poli.

Generare idee: crollato. Oggi non costa niente.
Giudizio convergente, cioè quale delle venti: merce rara.
Verifica, cioè distinguere il vero dal verosimile: merce rara.
Presenza fisica: l'ultimo fossato che non si può simulare.

Chi perde davvero sta nel mezzo, uno per parte.

L'estroverso generalista, il cui valore era essere brillante, avere idee a getto e conoscere gente. Tutte e tre sono ormai commodity.

E l'esperto di nicchia introverso, il cui fossato era sapere una cosa rara e complicata — cioè la cosa più replicabile che esista. Questo secondo caso viene omesso da quasi tutte le analisi, perché "vincono gli introversi" è una storia più comoda.

Il vantaggio degli introversi è reale. Ed è transitorio: esiste perché il canale dominante oggi è testuale e asincrono. Si assottiglierà quando gli assistenti diventeranno vocali e sempre presenti.

L'articolo completo di Roger Giuffrè, compreso il motivo per cui un ENTJ è oggi messo meglio di un INTJ nonostante guidi con la funzione che si sta svalutando:

Everyone has read the same take by now: remote work and AI are the introvert's revenge. The quiet ones finally win.

18/07/2026

An update from WebRobot.

We've validated automated setup of dedicated, single-tenant clusters — a customer's own environment, with distributed storage, our full data + AI platform, and automatic monitoring, all isolated per tenant. Two network modes tested, including one that scales elastically when the workload grows.

Next, we're building the full management experience into the dashboard — set up, monitor and manage everything from one place — and preparing white-label for agencies, so partners can offer this to their own clients under their own brand.

More soon.

13/07/2026

From data to trading strategies, on one platform 📊⚡

Two intense days of R&D: we integrated NautilusTrader into the WebRobot platform and turned our data into a full quantitative trading engine.

How does it work? The data WebRobot collects and analyses — sentiment, on-chain signals, trends — becomes the signals that drive trading and portfolio-management strategies. And you can define them simply and declaratively, without writing complex code.

The part we're most excited about: you can choose deterministic strategies (precise, reproducible rules) or agentic ones, where an AI agent decides the allocation, reasoning like a portfolio manager. 🤖

And it's not just theory: we tested everything on real market data and took a strategy all the way to an executed order in a sandbox environment — from data to trade. ✅

Another step toward our goal: making data instantly useful and actionable. 🚀

11/07/2026

🚀 Big milestone: the whole WebRobot platform now runs on YOUR own cloud.

We've deployed the full stack — data pipelines, AI agents, the in-app assistant, the data layer — end-to-end on a dedicated, single-tenant cluster. Your data never leaves your perimeter.

The hard part of doing this properly? Protecting our own technology when it runs on infrastructure the customer controls. That's exactly what we're locking down right now: our engine ships encrypted and activated by a revocable, per-tenant license — sovereign for you, protected for us.

Still on the list before launch: rolling that IP-protection layer across every component, plus one-click networking & certificates per tenant.

The next update drops very soon — stay tuned. 👀

05/07/2026

🤖 Give your AI agent a real browser — to analyze pages and set up data pipelines.

We just opened a free WebRobot MCP endpoint. Add it to Claude (Desktop or Code) — or any MCP-compatible agent — and it opens live pages in an anti-block Camoufox browser: your agent inspects the rendered page, infers the right selectors and fields, and builds + validates a working extraction pipeline. The analysis and setup, done conversationally — not by hand.

Point your agent at a URL and get to a validated data pipeline. No code, no signup.

→ Endpoint: https://mcp.webrobot.eu/mcp
→ webrobot.eu

{"jsonrpc":"2.0","id":"server-error","error":{"code":-32600,"message":"Not Acceptable: Client must accept text/event-stream"}}

29/06/2026

Final stress test: we didn't run a script — we gave an AI agent a goal. 🤖

"Scrape the whole catalogue, structure it, analyse it." Then we stepped back.

The agent designed the data pipeline itself, tested it small, scaled it to the full dataset on distributed Spark, checked its own work, and passed clean data to analysis & reporting agents — autonomously.

And recurring jobs reuse the validated pipeline directly, so they run fast and efficiently every time.

You bring the goal. The infrastructure builds the data product. ✅

28/06/2026

🤖 Our AI data-engineering agent just got a serious robustness upgrade.

Give it a goal in plain English — "scrape this site and compute the averages" — and it builds the pipeline, runs it, and returns the data. On its own.

Over the last two days we hardened it end-to-end:

🔹 Heavy jobs now run on the customer's own cloud (Bring-Your-Own-Cloud), scaling onto their VMs and cleaning up automatically.
🔹 The engine now tolerates the AI's natural output instead of failing on it.
🔹 A new validation gate catches mistakes before any compute runs — and the agent corrects itself.
🔹 We tracked a tricky "it scraped the wrong page" bug down to a single-line fix by testing every piece in isolation.

The result: a smarter, self-correcting platform that turns natural-language goals into real datasets — reliably. 🚀

25/06/2026

An AI agent ran a crypto portfolio for ~3 years. +126% — with 1/3 the drawdown of buy-and-hold. 🤖📊

We let a Claude-powered agent pick from 10 crypto chains each week, driven by on-chain data (DefiLlama TVL = where real capital is flowing) + market sentiment (Fear & Greed, used contrarian). Strict backtest, no future-peeking.

The result wasn't just the return (+125.7% vs +30% buy-and-hold) — it was the risk: max drawdown −24% vs −72%. The agent's superpower was patience: it sat in cash through three crashes and protected the capital.

Heads-up: here the 10-coin list was fixed by us — the agent chose within it. Next up: a dynamic universe that monitors fundamentals and adds new tokens as they launch, so the agent picks from a living list.

This is what WebRobot builds: AI agents reasoning over live, structured data — every decision explained and auditable.

Research backtest, not financial advice.

23/06/2026

🛰️ R&D update — WebRobot now enriches any dataset with free public data, in one pipeline stage.

We just shipped (and open-sourced, MIT) a data-enrichment plugin for the WebRobot distributed ETL. Drop a stage into any pipeline and every row is automatically augmented with authoritative, key-less public data — no API keys, no glue code.

It already works — real numbers, not slides. Six sources validated end-to-end on the cluster:

🌍 World Bank → Italy population 58,952,704, GDP/capita €40,385, inflation 0.98% (2024)
🇪🇺 Eurostat → Italy GDP €2.26 trillion, Germany €4.47 trillion (2025)
🏛️ OECD → per-country macro indicators (ITA / DEU / USA), one stage
🏠 HM Land Registry → "M1 1AE": 41 sales, median £138k (UK sold-price comparables)
📍 OpenStreetMap geocoding → "Colosseo, Roma" → 41.890, 12.491 · Roma · Lazio
🍽️ OpenStreetMap Overpass → 75 restaurants within 600 m of the Colosseo

Why it matters: enrichment is the unglamorous 80% of every data/AI project — joining your data to the world's. We turned it into a one-line, distributed, reusable building block: macro indicators for risk & market intel, sold-price comparables for real-estate arbitrage, geocoding to tie addresses to regions, nearby-POI scoring for location intelligence. Run it on six rows or six million — same Spark pipeline.

Open & extensible: a standalone, dependency-free SDK module, MIT-licensed. The same SDK lets anyone ship their own enrichment stage. EU-sovereign, running entirely on our distributed substrate.

🔗 Open source: https://github.com/WebRobot-Ltd/webrobot-enrichment-plugin

Contribute to WebRobot-Ltd/webrobot-enrichment-plugin development by creating an account on GitHub.

19/06/2026

🦊 R&D update — autonomous AI browsing, now on stealth Firefox

Today we shipped a major piece of WebRobot's autonomous data-extraction engine: our AI browser agent now drives stealth Firefox (Camoufox) over WebDriver-BiDi — going beyond the usual Chrome-only setups.

What's new and live:
✅ Full autonomy — the agent navigates, reads the page, clicks and types to complete real multi-step tasks (validated end-to-end on live e-commerce search and multi-site flows)
🌍 Geo-IP on demand — route any run through a residential exit IP in the country you choose, per request
🔍 No black box — every run now returns the exact selectors and the action trace behind each extracted value, so an AI run can be lifted into a deterministic, repeatable scraping pipeline. Our "autonomous scraper setup" thesis made real: agentic → deterministic, agnostic to Scrapy/Apify.
🔓 Open source — our Firefox/BiDi layer is an MIT hard-fork of browser-use (full credit to the upstream team).

This is the engine behind our conversational data-extraction demo.
👉 Try the chat demo: cloud.webrobot.eu