Artificially Intelligent, Actually Useful.

A Weekly AI Update So You Don’t Have To Wing It

AI image generation: Elsewhere

Oct 26, 2025 - 16:31 UTC

January 7, 2026

A Not-So-Artificial Review of AI in 2025

December 24, 2025

Artificially Intelligent, Actually Useful.

A Weekly AI Update So You Don't Have To Wing It

Prompting Was Just the Beginning

This week's newsletter is about a fundamental shift in how organizations are directing and deploying AI. Prompt-response is giving way to agentic engineering: coordinated AI systems that don't just respond, they act, and are governed by an orchestration layer that determines what runs, when, and with what oversight.

The articles this week trace that shift from concept to practice. AT&T routing 27 billion tokens daily through agent swarms but cut costs by 90%. Anthropic's agentic research shows 27% of AI-assisted work now includes tasks that were too expensive to do manually. The infrastructure is maturing.

You don't need to build these systems. But you do want to understand what they are, and where AI is heading.

Agentic engineering replaces vibe coding: developers now orchestrate AI agents connected to other agents and external systems, requiring formal oversight, not casual prompting.

AT&T routes 27 billion tokens daily through orchestrated small language models, cutting AI costs by up to 90% while boosting employee productivity.

Anthropic's research finds 27% of AI-assisted work now comprises tasks that weren't previously cost-effective to attempt. The developer's role is shifting from writing code to architecting systems.

Vibe Coding → Agentic Engineering

A year ago, “vibe coding” (generating software through natural language prompts) was a big story in AI development. Now, it's considered passé by some. Developers no longer need to write code most of the time. Instead, they orchestrate. Specialized agents research, build, and debug simultaneously, managed by platforms like Claude Code agent teams or Gas Town, and connected through protocols like Anthropic's Model Context Protocol (MCP).

You can think of agentic engineering as air traffic control for software development. But this type of agentic scale introduces real risk. A chatbot's wrong answer is an inconvenience. An autonomous agent writing flawed code into your production system is a major business liability.

Strategic Insight: Organizations treating agentic engineering as a prompting upgrade are missing the boat. The power is in the orchestration layer: how agents are coordinated, what protocols govern their communication, and where human oversight is truly needed in the workflow.

Read the Article →

AT&T Tripled AI Usage and Cut Costs By 90%

Processing 8 billion tokens a day and facing unsustainable costs, AT&T needed to make a significant change. Their solution: replace LLMs with an orchestrated stack of specialized small language models (SLMs), and they scaled to 27 billion daily tokens while cutting costs by up to 90%. But their chief data officer advises: don't make everything agentic for the sake of it. The real discipline is knowing where orchestration adds value and where it overcomplicates.

The Practical Angle: AT&T's architecture works because every agent has a narrow job and a human still oversees the chain. All actions are logged, data is isolated, and role-based access is enforced when agents hand off work. Orchestration without guardrails isn't scaling. It's risk.

View Full Case Study →

Anthropic's 2026 Agentic Coding Trends Report

The research finds that more than a quarter of AI-assisted work now includes tasks that weren't previously cost-effective to attempt manually. Single coding agents are evolving into coordinated teams that work autonomously for days at a time. And while developers report using AI 60% of the time, they can only "fully delegate" 0-20% of their tasks. Anthropic calls this the collaboration paradox: using AI effectively still requires thoughtful setup, active supervision, and human judgment, especially for high-stakes work.

Strategic Takeaway: The developer's role is shifting from writing code to architecting systems and evaluating agent output. This is a change that demands new skills, new review processes, and a fundamentally different relationship between humans and the software they build.

View Full Report →

Quick Hits

.Foundations

Workflow, Agent, AI Orchestration

Workflows are linear but break if one step fails. AI agents reason and self-correct, but cost more and can vary in success. AI agent orchestration combines both: agents working inside a structured workflow, each handling one task and passing its output to the next. This guide explains AI orchestration and shows you the set-up basics.

.Video

What Are Orchestrator Agents?

This video explains the progression from AI assistants (DIY) to agents (Do-It-For-You) to the next layer: orchestrator agents. As organizations deploy more specialized agents, the result is often a fragmented ecosystem. Orchestrator agents act as supervisors, routing work to the right agent, coordinating execution, and de-siloing tools.

.Deep Dive

Defining How AI Agents Communicate

Open protocols are standardizing how AI agents connect, communicate, and interact with humans, and each other. MCP has become the “USB-C of AI integrations,” eliminating the custom integrations needed to connect every new tool and model. Adopting these standards now prevents vendor lock-in as orchestration scales.

Industry Developments

OpenAI Releases Agentic Coding Model With Principles

GPT-5.3-Codex is OpenAI's agentic coding model that achieves higher accuracy while consuming significantly fewer output tokens. The company is also requiring agents to stay within agreed scopes and prioritize reversible actions, signaling a formal commitment to governed autonomous coding.

Google's Nano Banana 2 Upgrades Features and Speed

Google released a significant image generation model upgrade with Nano Banana 2. The model generates production-ready visuals up to 4K resolution, maintains subject consistency, and renders accurate text, all with increased speed. Google is also pairing the model with SynthID watermarking and C2PA Content Credentials for provenance tracking.

Alright, that's a wrap for the week!

The throughline this week: AI is shifting from a tool you prompt to a workforce you can orchestrate. But the most successful deployments aren't just about autonomy and scale; they're about knowing where orchestration adds value and where it doesn't.

Anthropic's research reinforces the point: even as agents take on more work, engineers can only fully delegate a small fraction of tasks, and this requires new skills, new security frameworks, and new ways of thinking about how AI gets deployed.

The organizations that benefit most aren't the ones that deploy the most agents. They're the ones building the orchestration layer with the right guardrails, the right oversight, and a clear understanding of where human judgment still matters.

Thoughts?

Artificially Intelligent, Actually Useful. is brought to you by iS2 Digital
Your Software at the Speed of Change

Click here to subscribe, if you want to receive our newsletter directly.

Artificially Intelligent, Actually Useful.

From Promise to Proof

A Not-So-Artificial Review of AI in 2025

Correlation Not Comprehension

Confident Yet Uncertain

Artificially Intelligent, Actually Useful.

Vibe Coding → Agentic Engineering

AT&T Tripled AI Usage and Cut Costs By 90%

Anthropic's 2026 Agentic Coding Trends Report

Quick Hits

Industry Developments

Got a project in mind?
Tell us about it.

Artificially Intelligent, Actually Useful.

From Promise to Proof

A Not-So-Artificial Review of AI in 2025

Correlation Not Comprehension

Confident Yet Uncertain

Artificially Intelligent, Actually Useful.

Vibe Coding → Agentic Engineering

AT&T Tripled AI Usage and Cut Costs By 90%

Anthropic's 2026 Agentic Coding Trends Report

Quick Hits

Industry Developments

Better Business, Not Better Buzzwords

Got a project in mind?Tell us about it.

Got a project in mind?
Tell us about it.