The new moat in the agent era is being the tool agents reach for.
A coding agent doesn’t reinvent a database. It wires up Supabase.
The best devtools companies will make themselves obvious to agents: easy to find, easy to reason about, easy to wire up.
Devtools are entering a golden age, but only for companies that realize they’re selling to agents now, not just humans.
Excited to speak at @PyData London this Sunday!
I’ll be talking about data context for agents beyond SQL: video, sensors, files, metadata, and Python workflows.
10:15 · Hardwick Hub Conference Room
Come say hi!
A few observations of my own (developing open source (DataChain) and SaaS for it, obviously using AI a lot).
* Last 5%-10% of quality / performance are not equally important in all cases. I personally pay 10x more attention to open source (core) vs some UI SaaS features. So, in a lot of cases it is an "an incredible result." for only $350 :)
* Sometimes harness / setup matters (ideally it should be clear from the code, conventions, rules, etc, etc) that performance is paramount (or quality, or this is a distributed system and we pay attention to all possible edge cases). Different setups and different priorities (and thus different price) for different projects.
* It is not exactly clear what would have happened if we were to allowe it to run for longer (with a proper goal). And then it will become a question of - price (and probably in this case it is still way cheaper to run AI for many days if needed) and is it even possible at all now (including maintainability of the end result).
It is hard for me to judge how many people actually blindly trust bc of "psychosis and lacking systems understanding" these days, and how many actually make informed decisions based on some sense and context.
Tbh I would not be surprised if AI didn't change much - just highlighted / accelerated bad and good taste, lack of curiosity, etc, etc.
I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem.
As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)!
I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work.
It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results?
88ms => 1.5ms
150K allocs => ~500 allocs
Incredible right? Nope.
My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path.
This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput.
The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity.
Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.
Turns out "Claude Code over files in S3" quickly becomes "rebuild half the data warehouse stack" 🫠
Schemas, datasets, lineage, file refs, etc.
OpenAI's Data Agent post made us feel slightly less insane 😄
Read more: https://t.co/6rkGBT7hM7
Total token use as a measure of AI literacy is wrong headed.
In my experience, after some baseline, more token use is inversely correlated with competency using AI.
@terronk@julien_c probably security? (banks are conservative for a, well, good reason). Good banks do use AI and copilots though, including agentic .. and for quite a while. It would be usually something like MS provided via VS Code, etc.
OpenAI's data agent - how structured / SQL data done right:
https://t.co/uAyo9v5nqz
🎥🔊🖼️ Multimodal data is harder: schemas and lineage aren't explicit - they must be inferred from Python code.
The upside: a single language removes an entire layer of context and simplifies reasoning.
✨ True meaning lives in the code ✨
LLMs broke out once text data hit scale.
Neuro is entering its own scaling era - EEG, DICOM/NIfTI imaging, 3D-scans.
Guess which part breaks first 👀 The data stack.
https://t.co/RW8iqrRMlS
Are there SE benchmarks that also measure "simplicity" and / or design of the generated code? Almost all time now just goes into "keep it simple", "refactor", etc ... Generating some code that just works is not an issue anymore.
DBT + Fivetran 🚀
A huge milestone for the "modern data stack". Consolidation is on - who's next? Snowflake ❄️? Databricks 🔥?
But maybe that doesn’t even matter. The next wave is here: Multimodal data stack
It's not replacing the old one - it's for different users:
🤖 AI, not Analytics
🧠 Unstructured, not tabular
📂 Files, not tables
🐍 Python, not SQL
⚙️ Way more CPU/GPU-hungry 😅
Tabular data is just one modality - and whoever wins multimodality might own tabular too.
Such an exciting time to be in the front row of this race 🔥
@Wattenberger just an observation: anthropic models do this in general (create a lot of tests files and reports) ... interestingly, working with GPT-5 - completely different experience. Not saying it is better at the end - but workflow is very different
Today we're launching Tasklet — an AI agent for automating your business.
Unlike ChatGPT, @TaskletAI actually does the work for you: connecting to your tools, triggering automatically, and handling tasks while you sleep.
Is there already a "Jervis"-like app that summarizes updates from agents when you are not at your computer and can get voice inputs to then feed back to those agents?
AI isn't just about text and code. What about sounds, videos, and sensors? 🎧🎬🔬
I’ll be at @MLOpsWorld Summit (Oct 6-9 in Austin, TX) sharing how to query inside the file ⚡️
Come nerd out with me in Texas 👋🤠
#MLOpsWorld2025
Look, say what you will about it, but right click editing a PHP file in an FTP client with upload-on-save is still the tightest and fastest feedback loop I've ever had in my life. We actually don't know how to do this anymore as an industry.