Rob Van Den Bergh

My preferred definition of "ARR" is your single-highest grossing minute of the year, times 525,600

Cluely is not the only ai company that's lying about ARR. You really think all these companies went to $100M in real fucking ARR overnight? Easy come, easy go.

Evolution of ARR: 1. Subscription revenue for the last 12 months 2. ⁠Projected subscription revenue for the next 12 months 3. ⁠Multiply your best month by 12 4. ⁠Multiply your best week by 52 5. ⁠Give out huge discounts but still count it as full revenue, do #4 or #5 to arrive at the final ARR, whichever is the highest. Celebrate like you did #1

@PavolRusnak Pretty cool prompt (Bristol)

Nice Nano Banana Pro prompt for weather app: CITY=Prague,Czechia Present a clear, 45° top-down isometric miniature 3D cartoon scene of [CITY], featuring its most iconic landmarks and architectural elements. Use soft, refined textures with realistic PBR materials and gentle, lifelike lighting and shadows. Integrate the current weather conditions directly into the city environment to create an immersive atmospheric mood. Use a clean, minimalistic composition with a soft, solid-colored background. At the top-center, place the title “[CITY]” in large bold text, a prominent weather icon beneath it, then the date (small text) and temperature (medium text). All text must be centered with consistent spacing, and may subtly overlap the tops of the buildings. Square 1080x1080 dimension.

@jspujji want

A 21 year old marketer is getting me 10M+ social impressions a year. Using a simple 3 step AI process: He: 1/ Compiles daily transcripts from my meetings 2/ Fine-tuned an AI model to convert them to content 3/ Tracks each metric obsessively I asked him to create a 5 min Loom so all of my portfolio companies can copy this process. Want the loom? Comment “want” to get it.

@hkanji Finally!

The mustache seems to be making a comeback npr.org/2025/07/01/nx-…

@Ubunta How does it go about how the data team define the company metrics?

@bob_bergh MCP takes care that, I created fine grained MCP tools to minimize

Building is believing. I created an MCP Server for Data Warehouses, connected it to live production systems, and ran hundreds of queries to validate what I suspected: this is the future of Business Intelligence and data exploration. What I built: – Integrated Snowflake, ClickHouse, and PostgreSQL for comprehensive data access – Developed multiple configurations optimized for different use cases and user personas – Implemented robust data governance controls at every layer – Enabled genuine conversational data exploration and automated insight discovery Chat-driven data analysis isn't just a concept anymore – it's a working reality that transforms how we interact with our data warehouses.

Late to the party, but a masterclass

@javitoshi @0xhirugohan There are quite a few guardrails you can put in place that restricts the actions that LLMs can take when interfacing with your DB. The best way we've found to do this at Fluent is by introducing a semantic layer.

yesssss yes yes yes yes a fully customizable dashboard every dashboard module can be created with a simple prompt. - "new users today + compared with yesterday" - "weekly active users, compared with prev week" and every time you do a text2sql it can be saved as a module for your dashboard i think everyone should have something like this, to save time and gain insights + it's really fun to build!!

Are any products using text2sql in production? Seems like you could never prevent all possible jailbreaks.

@codyschneiderxx You can just use fluenthq.com for this

business idea for you ai sql generator has the descriptions of each db column saved person uses plain text to have ai write a query distribution is google performance max ads to anyone who has searched "metabase" or similar probably a $10M company gl hf

@deedydas Would disagree that this is a fair measure of text-to-SQL - Spider 2.0 is more about testing GUI control/how data agents can work across a number of tasks. Wouldn't use it to evaluate accuracy in this context.

Difficulties of using language models in enterprise text-to-SQL workloads Spider 2.0 benchmark proposed with 15% achieved with o1-preview. Impressive Qwen 72B showing her. 6/10

ICLR is a top ML conference. All 10k+ papers from 2025 are in Open Review. The top rated papers include: — Scaling LLM interpretability to GPT4 scale — Changing light source w consistent image — 100x faster diffusion models — A provable theory for LLM jailbreaks Thread... 1/10

@PaddyStobbs You’ve now put the idea in my marketing leads head haha, so give it a few months

@bob_bergh Yup. Looking forward to the Fluent gilet :)

A leading VC recently told me that there is a high correlation between portfolio companies that gave him branded clothing at an early stage in their journey and their ultimate success as a company. And it has thrown my mental models of startup building out the window.

5/5 The evolution of the BI role should prioritise high-impact work: spotting growth trends, focusing on data quality, and empowering teams to self-serve. #Data team's impact will exponentially increase as a result of the shift too.

4/5 Here’s where AI can actually help. The idea isn’t to replace BI analysts but to automate repetitive tasks so they can focus on real insights. With natural language tools, non-tech teams could answer their own questions, and turn BI from a bottleneck to an enabler.

Read an interesting Reddit thread this week on the evolution of the BI analyst role - it's not getting any easier for them, even with all the BI tools on offer. Feels like it's time the role evolved from data 'firefighting' to more insight enablement and governance. A thread ⬇️