←
semantic slayer153 posts
semantic slayer avatar

semantic slayer

@datumnite

co-founder @lightdash_devs | euro trucker in the metaverse | 174bpm

◷ Joined February 2022▧ 10 Photos and videos
semantic slayer avatar
semantic slayer@datumnite·9h

Lots of people saying code mode is the answer here but let’s think bigger! What if every vendor exposed signed URLs for an open lake format? Then clients could bring the compute.

Slop Join: joining data from multiple MCPs inside an agent context window. I actually see this a lot in GTM teams when they’re pulling marketing data together. We need a better approach for big data through MCP.

semantic slayer avatar
semantic slayer@datumnite·Nov 17

Slop Join: joining data from multiple MCPs inside an agent context window. I actually see this a lot in GTM teams when they’re pulling marketing data together. We need a better approach for big data through MCP.

semantic slayer avatar
semantic slayer@datumnite·Nov 15

You have to invest time to migrating your local dev to a reproducible env. Running coding agents locally is easy because it has root access to all your secrets, repos, clis, even your networking. Ours easier to get started locally but it’s an anti-pattern for a software factory.

the biggest reason more people aren’t using cloud agents is because there’s not an obvious and simple way to migrate your local environment to the cloud feels like there should be an open source tool from one of the coding harnesses that solves this

semantic slayer avatar
semantic slayer@datumnite·Nov 13

i'm really enjoying open models - but i don't love the devx of any coding agent (opencode, t3, etc.). so at lightdash we're building our own coding agent optimised for 3 things: 1) intelligence like electricity - using @OpenRouter to allow you to pick any model / provider under the sun. 2) cloud-only - remote agents are the future because they can work non-stop and scale infinitely 3) sane defaults - workload identity, fast built in browser, shareable projects + config watch this space !

semantic slayer avatar
semantic slayer@datumnite·Nov 13

what do all these sentences have in common? the job bakes the metrics were laminated where the user goes it doesn't travel, it lands on it i asked opus 55 and it knew its own bullshit "guess: you're seeing these in agent output. models reach for this register to sound concrete when the thought underneath isn’t pinned down" But still love how it 'reaches for registers and pins down'

semantic slayer avatar
semantic slayer@datumnite·Nov 12

The future of analytics is not BI or notebooks. It's something in the middle. At Lightdash we're building "Composer Queries" these are small DAGs (like mini notebooks) that can chain together TB-level warehouse queries, duckdb execution, and python execution. Both dashboard building + notebook building were optimised for humans. We see composer queries as the ideal building blocks for agents. I've been interested to see tools in the data space decomposing notebooks into smaller execution graphs. While at lightdash we're evolving from single warehouse queries up to larger execution graphs. Meeting somewhere in the middle. That gives me a lot of conviction on this direction. Agents really want to work with data differently and I suspect a lot of exciting movement over the coming 2-3 months. It's a great time to be a data person!

semantic slayer avatar
semantic slayer@datumnite·Nov 12

i’m shook by fable 5.2 speed - a good reminder that a lot of tasks the intelligence is good enough and you should be maxxing speed

semantic slayer avatar
semantic slayer@datumnite·Nov 11

London, we're opening the Lightdash Loft! The @lightdash_devs team has built a space specifically for the data community to host events. Tomorrow, we're hosting our first event at the Loft - a happy hour with Joe Reis and Orchestra. Spots are nearly finished, so grab one here: luma.com/borj41tg And if you want to host an event at the Loft, get in touch!

Post attachment
semantic slayer avatar
semantic slayer@datumnite·Nov 9

Opus 5.5 at launch was so delicious. But i'm biasing to open-weight models for stability + control. Its hard to build automated products on top of models if they can alter behaviour overnight without a version change.

semantic slayer avatar
semantic slayer@datumnite·Nov 6

Velocity is everything. But not velocity out of fear or fomo. I see so many teams plagued by this and they're losing the capability of critical thought and they're burning out.

semantic slayer avatar
semantic slayer@datumnite·Nov 5

If you're trying to build frictionless agent experiences you have to keep an eye on the OAuth standards. Already huge extensions to OAuth have shipped (cimd, id-jag, act semantics) with a whole load more drafts that are work in progress. Most of the new standards you see posted online (like @WorkOS auth.md) is essentially a vendor layer on top of these standards. If you're building software you need to get to know! Delegating our credentials to agents is absolutely needed for a secure + frictionless experience going forward

semantic slayer avatar
semantic slayer@datumnite·Nov 4

London data folk! Next Thursday, we're hosting a low-key data happy hour with Joe Reis - which will also be our first event at the Lightdash Loft 👀 If you're in London, sign up here: luma.com/borj41tg 🚀

Post attachment
semantic slayer avatar
semantic slayer@datumnite·Nov 3

As of this week I’ve been coding exclusively with GLM 5.3 on @wafer_ai with @pidotdev . We’re nearly there!!! Free me from the shackles of the warped tokenomics we got from anthropic and openai

semantic slayer avatar
semantic slayer@datumnite·Oct 29

There is a reproducibility crisis happening in data right now and I want to talk about it! I'm joining @meltanodata and incident.io tomorrow to talk about how we can get trustworthy answers to our data questions using AI and how putting a chatbot on your warehouse just doesn't cut it. Bring me your toughest questions! 👉 Save your seat here: linkedin.com/events/75168272…