PM at JetBrains AI. Memes as a service š @ lananovikova @techhub.social š¦@ lananovikova .bsky.social Opinions are my own
Amsterdam
Joined February 2010
- Tweets1.9K
- Following863
- Followers610
- Likes484
The role of the context management and code base search in the token optimisation is unquestionable ā JetBrains Context is our response to this challenge. Please try it out
Coding agents shouldn't have to rediscover your codebase for every task.
JetBrains Context gives them code context to improve outcomes and cut costs on production-scale codebases.
Available now in early access as part of the JetBrains AI for Teams and Organizations rollout.
Learn more: jb.gg/u0chen
"How do we add an LLM here?" is the wrong question. The better one: "Why would an LLM beat a simpler solution for this?" A thread on telling real AI features from AI theater š§µ
"Full cost" = latency, infra, labeling, monitoring, legal exposure. Not just the API bill. None of this is glamorous. It's the gap between an AI feature that looks impressive and one you'd actually let near real users.
I wrote the full version for Roots ā it's in Russian š rootsproduct.ru/article/ai-pā¦ ā What's the best "AI feature" you've seen that a short list of canned answers would've handled better?
Everyone's adding observability to their agents. Almost no one is making it readable to them.Scattered signals, inconsistent naming, a dozen haystacks ā humans struggle with that. Agents can't reason over it at all.The AI conversation forgot the observability part.
#platformcon
@opentelemetry just graduated ā now the 2nd biggest CNCF project after Kubernetes, in production at 49% of companies. It's quietly becoming the context layer your production AI agents run on. Not a monitoring tool. A reasoning substrate.
#platformcon
Feeling faster isn't going faster.
The AI gains your devs feel aren't reaching production ā because 80% of the work breaks at the handoffs: tests, release, team-to-team. Generating code faster just floods a broken pipeline.
Kouros Aliabadi provided real numbers from production deployments of agents
#platformcon
Access is not adoption.
Booking.com gave 3,000+ devs every model, provider and tool ā adoption stalled. A platform to discover, share and scale patterns took daily use to 70%. Those devs now ship 30% more code.
Platforms create adoption. Buffets don't.
#platformcon