geek, scribe, coffee snob, and wanna-be cyclist. Contributor to Apache Spark and Delta Lake maintainer. Developer Relations at Databricks (opinions r my own)
Seattle
- Tweets 19,066
- Following 1,931
- Followers 2,950
- Likes 27,021
dennylee retweeted
At Databricks, we want employees using the best models on Day 1. That means moving quickly when models like Opus 5.5 and GPT-6 Sol launch, while evaluating real-world usage and managing the cost impact across thousands of employees.
Our AI engineering team uses Unity Gateway to manage access, spend and model selection at scale, then determines which models belong in our AI stack.
Learn how we roll out frontier models across Databricks: databricks.com/blog/how-data…
2
13
881
dennylee retweeted
Joe Rogan: Why are you a Democrat?
@JamesTalarico: My mother met my birth father, who had a drinking problem that sometimes led to being violent. My mom decided right then and there that she was leaving. She found a little apartment in East Austin and worked double duty at a hotel to provide for us. She saw Texas Democrats like Ann Richards, people who fought for the little guy. That was the classic Democratic Party. I asked my mom what we were, and she was like, ‘We're Democrats, because Democrats fight for the people.’ Our historical legacy is the party that fights for the little guy. That's why I'm a Democrat.
1,123
3,798
27,555
1,139,840
dennylee retweeted
Really cool results with autoresearch that are going into our model serving stack!
We @databricks now rank #1 on NVIDIA’s SOL-ExecBench kernel leaderboard, across all 4 tracks.
We put GPT-6 Astra and Opus 5 in a self-hillclimbing loop, and the agents beat the top GPU kernel engineers.
Total toten spend: ~$70K, far less than what kernel engineers get paid.
One learning: GPT-6 Astra and Opus 5 are still much better than OSS models at writing GPU kernels.
1
2
44
4,734
dennylee retweeted
AI agents need fast, accurate search at massive scale. Lakebase Search is now GA, bringing scalable vector and BM25 full-text search directly into Postgres.
- New frontier for price-performance, latency, and recall on VectorDBBench
- 4x cheaper than running pgvector for the same workload
- True pay-per-use with zero compute cost when idle
databricks.com/blog/lakebase…
6
6
41
3,560
Kimmel: Would you come to LA and be our mayor?
Mamdani: It’s three words for me always: New York City.
Aaron Rupar
38
428
4,306
99,668
dennylee retweeted
We @databricks now rank #1 on NVIDIA’s SOL-ExecBench kernel leaderboard, across all 4 tracks.
We put GPT-6 Astra and Opus 5 in a self-hillclimbing loop, and the agents beat the top GPU kernel engineers.
Total toten spend: ~$70K, far less than what kernel engineers get paid.
One learning: GPT-6 Astra and Opus 5 are still much better than OSS models at writing GPU kernels.
23
12
178
20,911
dennylee retweeted
We just tested the latest AI models on 2,400 engineers. The cost/quality frontier just shifted massively.
The TL;DR:
• Opus 5.5: Our new daily default - higher quality and 20% cheaper than prior baselines.
• GPT-6 Luna: Shockingly capable and 20x cheaper than Opus 5.5 for high-volume routing.
Must-read for anyone managing production AI workloads 👇
Crazy few weeks for model releases! Our findings @Databricks show several new models meaningfully advance the pareto frontier. Results below (online workload analysis of N=2,400 engineers, plus offline evals):
1. Two of the three models released last week clearly expand the cost/quality frontier: Opus 5.5 and GPT-6 Luna.
2. Opus 5.5 is now the highest quality mid-tier model. It is better than all prior Opus models, better than GPT-6 Sol, and better than GPT-5.6 Sol.
3. Opus 5.5 reduces same-task costs consistently by 20% in both offline and online analysis. This is against a baseline of Opus 4.8, the prior least-cost Opus model (Opus 5.0 was a bit of a dud with high costs and barely noticeable quality improvements).
4. Due to best-in-class quality and lower costs, Opus 5.5 is a strong candidate as an “every day default” model for coding, and we are now encouraging it for this purpose at Databricks.
5. GPT-6 Luna is very, very, very cheap. It was at least 20 times cheaper per-task than Opus 5.5 in every offline benchmark we tested and in observed online use.
6. GPT-6 Luna is surprisingly capable given how cheap it is. On one of our most difficult evaluation suites it roughly matches Opus 4.6 performance, while being 99.3% cheaper per-task than Opus 4.6 was at that time. That's a 100X cost reduction in ~9 months! This finding is preliminary and we are still evaluating Luna quality on a broader set of offline and online tests.
Our production setup: Unity Gateway to route workloads across models and trace agentic interactions. A mix of end-user harnesses including: Omingent (meta-harness), Claude Code, Codex, and Cursor.
22
33
391
77,781
dennylee retweeted
We've moved to an approach @databricks that gives "Day 1" access to 12,000 employees whenever a new frontier model is released. Very tricky to do while avoiding massive cost spikes and quality regressions. We've written up our approach if others are interested (see below).
2
2
27
1,277
dennylee retweeted
On it. We just allocated $750,000 to replace the roof at Engine 70.
Mr. Mayor! The team at @FDNY Engine 70/Ladder 53 on City Island are a lifeline for local communities — but their station roof is falling apart! We need to find $750K FAST to fix it up, so they can get back to protecting the East Bronx. Any chance you can help us out?
1,388
3,458
69,375
8,565,704
dennylee retweeted
.@AnthropicAI's Claude Sonnet 5.5 is now available on Databricks across AWS, Azure and GCP, governed by Unity Gateway.
It improves efficiency over Sonnet 5 for coding and agentic use cases, and reached Opus 5-level accuracy on document understanding, parsing and search. It joins Claude Opus 5.5, Claude Fable 5.1 and 60+ open-source and frontier models on Databricks.
Build domain-specific agents with Agent Bricks, deploy them as Databricks Apps with Lakebase-powered memory, and govern every call through Unity Gateway.
See documentation: docs.databricks.com/aws/en/r…
3
5
35
3,055
dennylee retweeted
Crazy few weeks for model releases! Our findings @Databricks show several new models meaningfully advance the pareto frontier. Results below (online workload analysis of N=2,400 engineers, plus offline evals):
1. Two of the three models released last week clearly expand the cost/quality frontier: Opus 5.5 and GPT-6 Luna.
2. Opus 5.5 is now the highest quality mid-tier model. It is better than all prior Opus models, better than GPT-6 Sol, and better than GPT-5.6 Sol.
3. Opus 5.5 reduces same-task costs consistently by 20% in both offline and online analysis. This is against a baseline of Opus 4.8, the prior least-cost Opus model (Opus 5.0 was a bit of a dud with high costs and barely noticeable quality improvements).
4. Due to best-in-class quality and lower costs, Opus 5.5 is a strong candidate as an “every day default” model for coding, and we are now encouraging it for this purpose at Databricks.
5. GPT-6 Luna is very, very, very cheap. It was at least 20 times cheaper per-task than Opus 5.5 in every offline benchmark we tested and in observed online use.
6. GPT-6 Luna is surprisingly capable given how cheap it is. On one of our most difficult evaluation suites it roughly matches Opus 4.6 performance, while being 99.3% cheaper per-task than Opus 4.6 was at that time. That's a 100X cost reduction in ~9 months! This finding is preliminary and we are still evaluating Luna quality on a broader set of offline and online tests.
Our production setup: Unity Gateway to route workloads across models and trace agentic interactions. A mix of end-user harnesses including: Omingent (meta-harness), Claude Code, Codex, and Cursor.
17
34
194
94,782
dennylee retweeted
What is a meta-harness and why do coding agents need one? 👇
Elise Gonzales joins @codebasicshub to explain how Omnigent provides a shared layer across coding harnesses, making it easier to switch and compose harnesses, apply guardrails outside the agent, share context across workflows, and collaborate through agentic pair programming.
🎥 Watch the full conversation: piped.video/watch?v=FQDP0Wr6…
#Omnigent #AIAgents #OpenSource #CodingAgents @dpcodebasics
2
7
693
dennylee retweeted
We found and eliminated an estimated $1.2M a year in wasted AI spend and lost productivity in one hour.
Here's how 👇
Using Unity Gateway tracing and Genie One, we identified seven small MCP-server bugs that were quietly driving:
• ~$499K/year in wasted tokens
• ~12,000 engineering hours/year in agent wait time
• 1,409 tool errors in a single 24-hour window
Genie One turned the trace data into a ranked list of what to fix, and a coding agent helped close the loop in an hour. databricks.com/blog/how-we-e…
13
13
78
6,189
dennylee retweeted
Game ready.
ALT Sam's going to play and he's ready to go. - Head Coach Mike Macdonald
29
145
3,211
69,575
dennylee retweeted
Few things make me prouder in life than the culture we built at Algorithmia with @platypii
2
11
392
dennylee retweeted
DoorDash spent $1.4M trying to stop Zohran Mamdani from becoming mayor.
This week's historic $131.5M enforcement action against DoorDash for underpaying NYC delivery workers shows why.
Yet another reminder that the only thing that can beat organized money is organized people. theintercept.com/2026/09/23/…
427
6,943
25,195
420,542
dennylee retweeted
Neon will have realtime very soon
I’m very excited about this. Coming next to the Neon backend: realtime.
Sync is incredibly hard to get right - especially if the goal is to make it great to work with, support as many database query use cases as possible, and scale well.
Sync coming to Neon was probably obvious to many after our announcement that the Electric team is joining Databricks but we can finally announce it officially today!
The Electric team is building sync for Neon. I asked @thruflo about joining Databricks and their first few weeks at the company: neon.com/blog/electrifying-n…
9
3
61
10,418