Telkom Indonesia rolls out agentic AI platform

TLDR: Big telco, big integrator, and the Pentagon all shipped serious agentic orchestration layers today, which is a pretty loud signal that "agents" are leaving the lab.

Top Signals

Telkom Indonesia launches BigBox agentic AI platform for enterprises

Telkom Indonesia introduced Agentic AI by BigBox, an enterprise platform for cross-system workflow orchestration, data integration, and end-to-end AI capabilities as of 2026-04-29. Internally, Telkom Indonesia is using the system to consolidate IT across its many entities, optimize operational costs and capital investment, accelerate data-driven decisions, and automate strategic business processes.

For you, this looks like a national telco betting that agentic workflows are the right abstraction for messy legacy estates. The platform sits on top of existing systems instead of ripping them out, so it matters if you are trying to modernize B2B operations in regulated or infrastructure-heavy environments. Details on architecture, models, and governance are still light, so treat this as a directionally important signal rather than a reference design.

Pentagon adds Google Gemini based agents to GenAI.mil

The United States Department of Defense added Google Cloud Agent Designer and Google Gemini models to the GenAI.mil platform, and internal users have already "vibe coded" thousands of agentic AI agents as of 2026-04-29. These agents are autonomous systems that use large language models to perform multi step tasks without human approval at each turn.

For anyone building serious production agents, this is a strong signal that defense workloads are moving beyond chatbots into orchestration and tool use. It also raises the security and safety bar: you should expect more scrutiny on auditability, model routing, and kill switches when the reference customer is the Pentagon. Details on evaluation, guardrails, and mission critical use cases are still sparse, so do not over extrapolate from early numbers.

DXC launches OASIS, an agentic orchestration layer for managed services

DXC Technology announced DXC OASIS, an open agentic orchestration layer that sits across existing tools to unify data, workflows, and systems into one intelligent operating model. DXC OASIS promises a trusted enterprise wide view of technology operations where every action is traceable and every insight is explainable.

This is very directly aimed at teams trying to tame sprawling ITSM, observability, and runbook stacks with agents instead of more dashboards. If you are building internal platforms, note the positioning: do not replace tools, orchestrate them and capture their events and decisions in a shared fabric. There are no public benchmarks or deep architecture diagrams yet, so you will need to map these claims to your own platform and compliance needs.

Introducing NVIDIA Nemotron 3 Nano Omni: Long-Context Multimodal Intelligence for Documents, Audio and Video Agents Nemotron 3 Nano Omni targets on-device, long context, multimodal agents that need to process documents, audio, and video in a single stack, which is interesting if you are building privacy sensitive or offline capable assistants.
Building AI-ready data: Vanguard’s Virtual Analyst journey Vanguard and Amazon Web Services outline eight principles for AI ready data, plus specific AWS services and measurable business outcomes, useful if your agents are blocked on messy enterprise data instead of model quality.
Migrating a text agent to a voice assistant with Amazon Nova 2 Sonic Amazon Web Services walks through architecture, design tradeoffs, and prompt changes for turning a text agent into a voice assistant using Amazon Nova 2 Sonic, with a focus on tools, sub agents, and latency pitfalls.
Run custom MCP proxies serverless on Amazon Bedrock AgentCore Runtime This guide shows how to host custom Model Context Protocol proxies on a serverless Bedrock AgentCore Runtime, giving you a programmable policy and observability layer for tools that must meet strict security requirements.
Show HN: Drive any macOS app in the background without stealing the cursor Cua lets you control macOS apps in the background without hijacking the active user session, which is exactly the kind of UI primitive you want for safe GUI operating agents; it is early but worth a look if you deploy desktop workers.
Show HN: Rocky – Rust SQL engine with branches, replay, column lineage Rocky is a Rust based SQL engine that bakes in data governance features like branching, replay, and column lineage; that is attractive if your agents modify or analyze production data and you need tight audit trails.
Why Codex works better than Claude Code for my production monolith A practitioner compares Codex and Claude Code on a long lived Python backend and still prefers Codex, a reminder that real world tooling choices depend heavily on architecture and workflow rather than benchmarks.
Choco automates food distribution with AI agents OpenAI profiles how Choco uses AI agents to streamline restaurant supply chains, showing concrete productivity gains and hinting at design patterns for logistics heavy agent deployments.
Cybersecurity in the Intelligence Age OpenAI lays out a five part plan for AI powered cyber defense and critical system protection as of 2026-04-29, which is relevant if your agents interact with sensitive infrastructure.
Speech translation in Google Meet is now rolling out to mobile devices Simon Willison reports on Google Meet’s mobile speech translation, which kind of works already and previews user expectations for real time multimodal assistants.
Native Deployment Checks are now available Vercel can now run lint and typecheck on every deployment in parallel with builds, giving you another place to wire in static analysis or agent generated diff checks before changes hit production.
[AINews] ImageGen is on the Path to AGI Latent Space reflects on the ongoing GPT Image 2 wave and what increasingly capable image models might imply for broader AGI progress and multimodal agent design.