dTelecom Makes Voice AI Agents a 10-Minute Build

dTelecom Makes Voice AI Agents a 10-Minute Build

A Solana-backed startup called dTelecom is claiming developers can build complete voice AI agents — including speech-to-text, LLM processing, and text-to-speech loops — in roughly 10 minutes [4]. Their decentralized real-time communication network promises low-cost, low-latency infrastructure for voice, video, and AI workloads.

The company just launched Phase 2 of a $2.6 million airdrop campaign, backed by Solana Foundation, Google, and peaq [5]. They're positioning themselves as a challenge to the $3.5 trillion traditional telecom market by democratizing real-time AI communication infrastructure.

If the 10-minute claim holds up, this represents exactly the kind of abstraction layer that turns complex AI orchestration into simple configuration — a key trend in the post-code era.

Transformers Learn to Execute Code Internally

Percepta Research published findings showing that Transformer architectures can simulate Turing machines internally, enabling complex computations without external Python interpreters [6]. This builds on theoretical work proving that constant bit-size Transformers are Turing complete, but now demonstrates practical execution.

The research, published three days ago and actively discussed on Hacker News, suggests we're moving toward unified neural architectures that handle reasoning, coding, and execution in closed loops [7]. Instead of LLMs calling out to external tools, they could perform computations directly within their own forward passes.

This points toward a future where the boundary between "thinking" and "computing" dissolves entirely within AI systems.

US Withdraws Global AI Chip Export Controls

The Commerce Department quietly withdrew its planned rule for sweeping global AI chip export controls, according to government website updates from yesterday [8][9]. The move removes licensing hurdles that would have affected Nvidia, AMD, and other semiconductor companies, but creates new regulatory uncertainty.

The withdrawal eliminates what many saw as regulatory overhang on AI hardware companies, though it leaves the broader question of AI export policy unresolved [10]. For companies building AI infrastructure, this removes one potential compliance burden while highlighting the fluid nature of AI governance.