Tech Pulse Daily: June 25, 2026
Curated by Dillip Chowdary • Morning edition, IST
Today's Top Highlights
- 🌶️OpenAI and Broadcom unveiled a custom LLM inference accelerator after a nine-month tape-out cycle.
- 🤖OpenAI says 80.6% of sampled individual Codex users delegated at least one task over 30 minutes of human work.
- 💬Anthropic is putting Claude directly into Slack channels with selected tool, data, and codebase access.
- 🧰Seven validated agentic AI solutions from four partners are now listed in AWS Marketplace.
- 🛡️National Academies warns agentic systems are changing both attack automation and adaptive defense.
OpenAI + Broadcom: Jalapeño Inference Chip
OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom Intelligence Processor for LLM inference. The system is aimed at lower-latency, higher-efficiency serving for ChatGPT, Codex, API workloads, and future agentic products.
- OpenAI says its models helped accelerate the chip design and optimization cycle.
- Engineering samples are running ML workloads at target production frequency and power.
- The companies describe a multi-generation platform planned for gigawatt-scale deployments.
- Lower inference cost can show up as longer agent runs, cheaper API calls, and steadier peak-hour access.
OpenAI: Codex Moves Beyond Developers
OpenAI published an economic research update showing how Codex usage has shifted from short coding prompts to long-running delegated work. The clearest signal is that non-developers are now using agents for automation, data transformation, tooling, debugging, and structured analysis.
- By May 2026, 80.6% of sampled individual users made one request estimated above 30 minutes of human effort.
- 70.2% made at least one request estimated above one hour of human work.
- OpenAI says average worker usage now routes more than 85% of output tokens through Codex.
- Top internal users generated more than 60 hours of Codex agent turns per day by June 2026.
Anthropic: Claude Tag Enters Slack
Anthropic introduced Claude Tag, a Slack-first workflow where teams can mention @Claude inside selected channels. The product direction matters because it turns enterprise chat into an agent workspace with memory, scheduled planning, and access to approved tools.
- Claude can build context from selected Slack channels instead of relying on copied prompt snippets.
- Teams can connect data, tools, and codebases according to their own access boundaries.
- Anthropic frames the feature around tasks Claude can plan and complete later.
- Admins should review channel permissions because chat context becomes operational agent memory.
GitHub Copilot: Auto Model Routing for Free + Student
GitHub changed Copilot Free and Copilot Student plans so auto model selection is the default and only model picker experience. The change nudges lower-tier users toward provider-managed model routing instead of manual model choice.
- Auto mode dynamically selects a model for each task across supported model families.
- Access still follows plan restrictions, so the feature is routing control rather than unlimited model access.
- Developer tools are moving model choice behind policy, latency, and task-fit heuristics.
- Document when auto mode is acceptable and when paid tiers need explicit model control.
AWS: Partner Agent Factory Lands in Marketplace
AWS published the latest GenAIIC Partner Agent Factory cohort in AWS Marketplace. The catalog approach is important because buyers can evaluate agent solutions through existing procurement, billing, and enterprise spend commitments.
- The cohort covers document translation, video intelligence, governance, service, finance, retail, and clinical documentation.
- MegazoneCloud, Nubiral, Loka, and NorthBay are included.
- AWS names Amazon Bedrock, Bedrock AgentCore, and Strands Agents as core building blocks.
- The strongest listings emphasize retrieval precision, audit trails, cost governance, and self-correction.
AI Cybersecurity: National Academies Rapid Consultation
The National Academies released a rapid expert consultation on how generative AI, large language models, and agentic systems are changing cybersecurity. The report frames AI as both an attacker accelerator and a route to continuous, adaptive defense.
- AI can speed vulnerability discovery, exploitation, deception, and operational scaling.
- AI-enabled tools can support more adaptive monitoring, triage, and mitigation workflows.
- The consultation calls out weak metrics for attacker-defender advantage and cyber resilience.
- Raise baseline software quality, incident response speed, and threat-intelligence sharing before agentic attacks scale.
Kubernetes: v1.37 Alpha 2 Release Window
The Kubernetes release calendar marks v1.37.0-alpha.2 for June 24, with code freeze scheduled for June 29. Platform teams testing Kubernetes 1.37 should treat this week as the handoff from exploratory feature work to stabilization planning.
- v1.37.0-alpha.2 sits in week six of the Kubernetes release schedule.
- Code freeze is listed for June 29, 2026, tightening merge discipline.
- Placeholder PR deadlines land in early July, so docs owners should start review now.
- Track deprecations, admission behavior, and scheduler changes before staging cluster upgrades.
This Week in Tech
National Academies AI cybersecurity release webinar replay and publication review.
Kubernetes v1.37 code freeze begins, shifting feature teams into stabilization.
Kubernetes docs placeholder PR deadline for the v1.37 release cycle.
Developer Resources
Key Takeaways
- 1Inference economics are becoming a strategic product lever as OpenAI moves deeper into custom silicon.
- 2Agent governance needs channel, codebase, and tool-access policies before Slack-native agents spread.
- 3Developer platforms are hiding model choice behind routing systems, so teams need explicit quality and cost checks.
- 4Marketplace agents are shifting from demos to procurement-ready products with audit and governance claims.
- 5AI cyber risk is now an operating-model problem, not only a tooling problem.
Market Snapshot
Currency Exchange
Source checked June 25, 2026 morning: BookMyForex live rate.