LAUNCHED: Bits Code
You've given coding agents access to your code, now you can give them access to what actually happened in production.
Take a Datadog finding (an error, performance issue, flaky test, vulnerability, or slow query) and turn it into a reviewable PR grounded in traces, logs, profiles, runtime context, and repository code.
see the docs 👉 docs.datadoghq.com/bits_ai/bits_a…
check the blog 👉 datadoghq.com/blog/bits-code/
Still managing long-lived API keys and secrets?
We've added more ways to use identity federation so your applications, services, and workloads can retrieve short-lived Datadog Developer credentials directly from your cloud environment, reducing secret management overhead and improving security.
Read all about it 👉 datadoghq.com/blog/datadog-a…@datadoghq
The bottleneck isn't coding anymore, it's designing the workflow, and the teams that move fastest won't be the ones with the most agents.
They'll be the ones that know exactly where humans should step in, where agents can operate autonomously, and how work flows from idea → roadmap → code → production.
"Productivity will be determined by can you get rid of the bottlenecks."
That's a very different future than "AI replaces developers."
The biggest productivity gains won't come from better models alone. They'll come from eliminating the bottlenecks between an idea and something shipping to production.
@datadoghq@warpdotdev@zachlloydtweets
"The future isn't one coding agent. It's a system of specialized agents." - @zachlloydtweets, CEO and Founder of @warpdotdev, on what building software with agents actually looks like:
→ Agents triage issues
→ Agents write specs
→ Agents implement code
→ Agents review code
→ Agents verify changes
@datadoghq
Feature flags are becoming part of the AI engineering workflow.
With @datadoghq's Feature Flagging MCP tools, agents can:
→ Create feature flags
→ Check implementations in code
→ Inspect flag configurations
→ Start canary rollouts
From validating fallback values to updating targeting rules, your agent can help manage feature releases without leaving the editor.
Watch the demo ↓
LAUNCHED: Two new ways to run experiments in Datadog 🧪
✅ Fixed Sample Experiments are now available, giving you another analysis method to choose from during experiment setup.
✅ Bring Your Own Flag (BYOF) is now generally available, making it easier to connect experiments to your existing flag infrastructure.
Whether you're looking for more flexibility in how you analyze experiments or want to experiment on top of your existing flag infrastructure, both are now available in @datadoghq.
Thinking about consolidating your feature flag tooling?
@datadoghq's feature flag migration CLI helps you migrate flags, map environments, validate targeting, compare evaluations side-by-side, and export migration reports so you can verify everything works before cutover.
See how it works ↓
We had a packed room for the Product Analytics session at DASH 👏🏻
Developers aren't just shipping features anymore, they're increasingly expected to understand how those features get adopted, where users get stuck, and what drives activation.
The interest in Product Analytics this week was a good reminder that building the product and understanding its impact are becoming part of the same workflow.
@datadoghq
"We rewrote our entire database system from Python to Rust." - @thsottiaux, @OpenAI
AI is making large-scale rewrites and modernization projects economically viable again. That's the pro.
And yes there are some cons sometimes - "One bad example is when it deleted an entire research cluster."
The conversation is no longer:
"Can AI write code?" it's "How much autonomy are you willing to give it?"
The more agents move from generating code to operating systems, the more important it becomes to understand:
→ What they did
→ Why they did it
→ Which tools they used
→ What caused failures
That's why observability is becoming a core part of the AI stack.
@datadoghq
"The biggest breakthroughs in AI coding aren't coming from bigger datasets, they're coming from feedback loops" - @_sholtodouglas of @AnthropicAI
Sholto described two broad ways models learn:
→ Pretraining (learn from existing data)
→ Reinforcement learning (try things, get feedback, improve)
Coding is particularly suited to the second approach, and that's why coding capabilities are improving so quickly.
You can measure success:
Did the code compile?
Did the tests pass?
Did the task get completed?
The model gets a signal and improves.
Sholto pointed to @cognition's latest FABLE evaluation, where Claude Opus 4 was reported to be 2–3x better than competing models at producing mergeable, high-quality commits on difficult coding tasks.
How often does a human need to step in?
Sholto said his personal expierence:
~30 second interventions
→ then ~3 minute interventions
→ now somewhere between 30 minutes and 3 hours
That's a massive change in a very short period of time.
A better way to think about progress might be:
How long can the model work autonomously before you need to take over?
#fable#anthropic@datadoghq@atalwalkar
Our very own @atalwalkar sat down with @AnthropicAI's @_sholtodouglas to chat everything AI at DASH - from the release of Fable to deep reasoning, clips incoming… 🔥
287 Followers 351 FollowingAmplify Japan User Group Organizer | AWS Community Builders | Datadog Champion | Top Japanese PRC for Adobe Illustrator 2019,2020,2022 | Opinions are my one
5K Followers 485 FollowingHigh Tech Girl in High Heels, SVP Eng @Aurora_inno, ex VP @Google, ex SVP/GM @VMware Determined to have a positive outlook on life. Tweets are my own.
6 Followers 79 FollowingAI Data Pipeline Engineer | Building RAG systems &
data pipelines for AI products | CS/Engineering student
| Documenting the build - Ghana 🇬🇭
440 Followers 379 FollowingNew Zealander, living in Switzerland. I brew, I cook, I run. Engineering Manager @datadoghq. Views expressed were mine at one point.
31K Followers 10K FollowingWe cover the Big Apple 24-7. We're here to celebrate all of the great restaurants, businesses, cultural events, stories and people that make NYC so unique.