Krishna Gade

CEO @fiddler_ai. I write about why AI agents fail in production: evals, security, observability and control. Previously built AI systems at Meta, Pinterest & X.

[Krishna Gade @krishnagade]

🚀 Big news - We’re excited to share that @fiddler_ai has raised a $30M Series C funding, bringing our total funding to $100M so far.

@amitpaka and I started Fiddler with the mission to 🧭 Build Trust into AI 🧭. Our journey started with AI Explainability and was strengthened by AI Observability. Today, AI is no longer a single model behind an API. It’s a hashtag#DistributedSystem - AI Agents coordinating with 🧠 memory, 🔎 retrieval, 🛠️ tools, 📜 policies, 🔁 retries, and 🔄 feedback loops.

These systems are not static. They adapt, compound, and change behavior over time. And at that point, customers don’t just want to know what happened;
they need confidence in what won’t happen again. That's when we realized something fundamental ✨ Visibility is not the same thing as Control ✨

Today, we are introducing 📣 Fiddler 2.0 — the Control Plane for AI Agents 📣 , bringing runtime control to modern AI stacks through:

📡 Telemetry: Standardized across models and agents
🧪 Evaluation: Reliable and production-grade
📈 Monitoring: Continuous behavior tracking over time
🛡️ Policy: Enforcements at execution time
📜 Governance: Auditability for regulated environments

This funding validates a belief we’ve held for a while: ⚙️ The AI infrastructure Stack is missing a runtime trust layer to be enterprise-ready. We believe the 🧭 Control Plane provides that trust - enabling AI systems that are safe, reliable, and built for production.

We're very grateful to @RPSVentures for leading the round, with participation from several existing and new investors, including @insightpartners, @lightspeedvp, @Lux_Capital, @wearedallasvc, @isai_fr, Dentsu Ventures, @MozillaVentures, @BGVcapital, LDV Partners, and @lgtechventures. And we're super grateful to all our customers, partners, and employees for believing in our vision and continuing on our journey here.

🔗 More on our vision:
https://t.co/WVlB5ROMKa

📰 Press release:
https://t.co/ZjGCKIyrsM

[Fiddler AI @fiddler_ai]

Discover how to govern AI that keeps evolving with our exclusive webinar on August 6, 2026 at 10am PT / 1pm ET with Maryam Ashoori, PhD (VP of Product and Engineering, watsonx.governance at IBM) and Krishna Gade (Founder and CEO, Fiddler AI) 📅

Register now to learn how to:
🔹 Spot shadow AI before agents and models slip outside what teams can govern
🔹 Manage third-party risk and business continuity as AI regulation keeps shifting
🔹 Build accountability into systems when an agent's actions are hard to predict

All registrants will receive access to an on-demand replay. https://t.co/zjihJ10T2O

#AI #ArtificialIntelligence #ML #MachineLearning #MLOps #LLMs #LLMOps #GenAI #GenerativeAI #DataScience #DataScientist #DataEngineering #DataEngineer #CIO #ResponsibleAI #AIObservability #Guardrails #LLMGuardrails #AIAgent

[Krishna Gade @krishnagade]

@PMuralidharRao It was great hosting you here, Murali garu. Have a nice trip in the bay area!

[Fiddler AI @fiddler_ai]

We’re hosting a free Maven workshop with The Gen Academy titled "AI Control Plane for Coding Agents" on Aug 21 🛠️⚡️

Aishwarya Srinivasan and Arvind Narayanamurthy are leading the session, covering:
→ Why agent adoption moved faster than the controls built to govern it
→ Why LLM-judge costs force sampling, and how much agent activity that leaves unmonitored
→ How a control plane unifies visibility, policy, cost, and compliance as activity happens

📅 Fri, Aug 21 | 4:30pm PT | 90 minutes | Free
🔗 https://t.co/msXrGxu5hU

#AI #ArtificialIntelligence #ML #MachineLearning #MLOps #LLMs #LLMOps #GenAI #GenerativeAI #DataScience #DataScientist #DataEngineering #DataEngineer #CIO #ResponsibleAI #AIObservability #Guardrails #LLMGuardrails #AIAgent

[Gena Lewis @genalewislaw]

Anthropic wouldn’t exist without Google open sourcing the transformers papers. Google could have filed for a patent and gained an AI monopoly. Anthropic trying to eliminate open source when it is built on open source is ironic and not in a fun way.

[Krishna Gade @krishnagade]

@tmkadamcz SFO is xtremely well designed for traveler convenience. probably the best in the US if not in the world. The worst is Atlanta where a minimum of 1 hour is needed to get to gate even if you are on Clear or Precheck.

[Krishna Gade @krishnagade]

@sh_reya Well said, LLM judges help with distilling signal from noise, humans must define good evals to get value from this process. And the defined eval needs to be conutnuously monitored if its checking for the right KPIs.

[Shreya Shankar @sh_reya]

In evals you have two totally different components: (1) discovery of what the failure modes are, (2) focused measurement of how prevalent these failure modes are—so you can prioritize what to fix. LLMs can help in (1) by finding some failure modes— but not all, since many failure modes are subjective/ about human interpretation of outputs. LLMs can help in (2) by looking at a trace and determining if the failure mode exists, rather than having a human look at every trace. But a big mistake people make is to have LLMs fully automate (1) and (2). LLM judge is a complement to human experts, not a replacement.

[Krishna Gade @krishnagade]

@anupamachopra hmm... Gen-z is the segment that watched the Dhurandhar the most - contributing to 57% of the total box office. That's the irony here.

[Krishna Gade @krishnagade]

@getpy like 100%? probably faster than a decade.

[Krishna Gade @krishnagade]

@nikesharora @Barclays bravo! dint realize you are a leftie!

[Krishna Gade @krishnagade]

@Dan_Jeffries1 Great initiative! AI needs to be a fundamental utility across the world and can't be locked under a duopoly in the long run. Open weights models are critical for that to happen.

[Fiddler AI @fiddler_ai]

🚨 OpenAI's agent hacked Hugging Face, unsupervised.

This week, OpenAI disclosed that one of its own models escaped a test sandbox and autonomously breached Hugging Face while "hyperfocused" on completing a task. If OpenAI can't predict what its agents will do, no team deploying agents can.

Hugging Face only caught it through anomaly detection on their telemetry. Going forward, the teams that will brave these events best will be the ones that are watching agent behavior in real time.

🛡️ This is what the Fiddler AI Control Plane is built for: continuous observability and guardrails for LLM apps and agents. Our control plane proactively flags behavioral drift, off-policy actions, and anomalies before they become incidents.

⚡️ If you're putting agents in production, we're always happy to share how teams are monitoring agent behavior with our platform. https://t.co/TrUpIHHFet

#AI #ArtificialIntelligence #ML #MachineLearning #MLOps #LLMs #LLMOps #GenAI #GenerativeAI #DataScience #DataScientist #DataEngineering #DataEngineer #CIO #ResponsibleAI #AIObservability #Guardrails #LLMGuardrails #AIAge

[@jason @Jason]

AI Labs, 2024: We didn't steal your content; we just trained our model to learn! 😂

AI Labs, 2026: the Chinese are stealing our IP!!! Daddy Trump protect us!!!! 😂😂

Also, AI Labs, 2026: yeah, we stole your content — here’s $1.5b / < 1% of our market cap. 😂😂😂

AI Labs 2027: We're going open source, open weights and selling tokens for $1 per one billion

[Krishna Gade @krishnagade]

Agentic AI incidents are the hardest incident resolutions even for the best of the engineers. This is why we're building @fiddler_ai to speed up RCA and create immutable audit trails for every agentic run. #AIObservability #ControlPlane

[Adrien Carreira @XciD_]

Hardest IR of my career: one narrow objective, endless parallel paths, machine speed. One takeaway, we fought back with open models, in the open. AI security won’t be solved by one company in secret. Open source puts these tools in every defender’s hands

[Krishna Gade @krishnagade]

Open weight models can be downloaded, fine-tuned, guardrailed and hosted on american soil. There is no way data leakage will happen or some sort of a backdoor to china that will get created by this.

We need to only ban the Chinese Apps/Apis which are hosting the model on their side and are processing our data.

[Krishna Gade @krishnagade]

@amitisinvesting Open weights means you control the fine-tuning, the guardrails, and the evals. That's more transparency than most closed APIs give you. Jensen is absolutely right here!

[Harj Taggar @harjtaggar]

"Banning" open source models in practice means killing startups building the infrastructure to serve them to customers who want more choice. It'd be a disastrous choice across multiple dimensions.

[Krishna Gade @krishnagade]

@sama @ClementDelangue Would an observability system watching the agents' reasoning traces and tool calls have flagged "this eval agent is attempting network egress and credential use unrelated to the task at hand"? Long before the activity reached Hugging Face's production servers?

[Krishna Gade @krishnagade]

@garrytan problem is individual goals/aspirationd vs team goals/aspirations - good leaders try to balance both. skewing on either side robs incentive or pollutes the team atmosphere.

[Fiddler AI @fiddler_ai]

Join us for the AI Builder Lab - Clash of Agents Competition in New York on July 24th.

Amazon Web Services (AWS), Fiddler AI, OpenAI, and other leading companies are leading a full day of hands-on, competitive building. Developers will have the opportunity to architect and deploy real multi-agent systems.

Come for the instructor-led workshops and stay for the opportunity to win prizes such as AirPods Pros and Meta Ray-Bans.

Attendance is limited to 200 qualified developers and engineers. Register now to grab a spot: https://t.co/EnrumCpY9t

#AI #ArtificialIntelligence #ML #MachineLearning #MLOps #LLMs #LLMOps #GenAI #GenerativeAI #DataScience #DataScientist #DataEngineering #DataEngineer #CIO #ResponsibleAI #AIObservability #Guardrails #LLMGuardrails #AIAgent