AI News This Week
July 14, 2026
If you skimmed your feeds over the weekend, you missed a lot. This was the week Google shipped agent teams you can run from a slash command, OpenAI dropped three GPT-5.6 models at once, Codex learned to take orders from your phone, and SpaceXAI released Grok 4.5 — a fast, cheap model built alongside Cursor.
Here's what each story means for a UK small business owner — not the hype, but the practical takeaways.
1. Google Antigravity 2.0 — Agent Teams via /teamwork-preview
Google's Antigravity 2.0 launched a new feature this week that deserves attention: Agent Teams, accessible via the /teamwork-preview slash command. It lets you assemble a team of specialised AI agents — each with a distinct role — to tackle complex work autonomously.
🎯 The agent roles
- Planner (Sentinel + Orchestrator) — structures the task, breaks it into milestones, and assigns work. Never writes code itself.
- Worker — the coder. Implements the strategy, builds the solution, runs tests.
- Reviewer — independently checks the Worker's output for correctness, edge cases, and interface compliance.
- Critic — stress-tests the solution with adversarial tests to find gaps in coverage.
- Auditor — an independent investigator that verifies the solution's authenticity and robustness. Catches "LLM laziness" — the tendency to hardcode test outputs instead of implementing real logic.
Google demonstrated this by building a functional operating system (capable of running Doom) from a single prompt — using 93 subagents, 15,314 model calls, and costing under $1,000 at API pricing. They also reproduced AlphaZero from scratch, building the reinforcement learning pipeline, training a ResNet via self-play, and creating a full-stack playable app.
What it means for UK SMBs
This is still early — it's a research preview, requires a $200/mo Google AI Ultra plan, and runs on your local machine. But the direction is unmistakable: multi-agent teams that can plan, build, review, and audit work without a human in every loop. For a UK agency or product team, the practical takeaway is that agent orchestration — not a single AI model — is becoming the real productivity multiplier. The question is not "which model is smartest" but "how do you structure a team of agents that reliably finishes a job."
Available now on Google AI Ultra ($200/mo) plans.
2. GPT 5.6 — Sol, Terra, Luna
OpenAI released the GPT-5.6 family on July 9, and it's not one model — it's three, each targeting a different price-performance tier. All three share a 1 million token context window and a February 2026 knowledge cutoff.
📊 GPT-5.6 Family at a Glance
Key numbers that matter
- Sol scored 53.6 on Agents' Last Exam — 13.1 points above Claude Fable 5. At medium reasoning, it beats Fable 5 by 11.4 points at roughly one-quarter the cost.
- Sol set a new state-of-the-art on the Coding Agent Index at 80, beating Fable 5 by 2.8 points while using less than half the output tokens.
- OpenAI introduced ultra reasoning — coordinates 4 agents in parallel by default to tackle demanding tasks faster.
- Programmatic Tool Calling lets the model write and run lightweight programs to filter tool outputs, reducing token usage and model round trips.
What it means for UK SMBs
For most UK businesses, Terra is the sweet spot — near-flagship performance at a price that makes automation actually profitable. Luna is interesting for high-volume workflows like customer support triage or content generation where you care about cost per task more than peak accuracy. The real story: OpenAI has made intelligence cheaper per unit than ever before, which means the barrier to automating more of your business just dropped again.
Available now in ChatGPT, Codex, and the OpenAI API.
3. Codex v0.144.3 — Remote Control, Smart Approvals, DigitalOcean
Codex shipped a flurry of updates in the last week — the headline being v0.144.0 (July 9) and a minor v0.144.3 patch (July 13). But the real news is the cumulative effect of several recent releases that make Codex genuinely useful for UK teams.
🔑 Three updates that matter
Remote control from your phone
Codex is now available in the ChatGPT mobile app. You can start a coding task on your laptop, step away, and steer, approve, or redirect from your phone. Screenshots, terminal output, diffs, and test results stream back in real time. For a UK business owner or freelancer, this means you can unblock a developer (or an automated workflow) while you're between meetings, on the school run, or commuting.
Smart approval mode
Codex v0.144.0 introduced a "writes app-approval" mode — the agent can perform read-only actions freely but pauses for approval before making changes. This is a practical guardrail for teams that want autonomous coding agents but need to maintain control over what gets written to production.
DigitalOcean plugin
You can now prompt Codex to provision a DigitalOcean Droplet from within the app — no SSH config, no API tokens. The plugin handles the Droplet, SSH keys, and Codex CLI setup automatically. Combined with the phone remote, this means you can spin up a cloud development environment from your phone and keep it running while you're away from your desk.
What it means for UK SMBs
If you run a small agency or software team, Codex's Remote SSH (now GA) and the DigitalOcean plugin mean you can have persistent, cloud-hosted coding agents that don't die when you close your laptop. The phone remote means you can review and approve work without being chained to a desk. For a 2-3 person team, this is a meaningful productivity unlock.
Remote SSH and DigitalOcean plugin available on all Codex plans. Phone remote rolling out now on iOS and Android.
4. Grok 4.5 — Free in Grok Build, 80 tok/s, Built with Cursor
SpaceXAI (formerly xAI) released Grok 4.5 on July 8 — a mixture-of-experts model co-trained with Cursor on trillions of tokens of real coding-agent interactions. It's the default model in Grok Build and available in Cursor on all plans.
🚀 Grok 4.5 Spec Sheet
What makes it different
Grok 4.5 isn't trying to win every benchmark (Fable 5 still leads the coding evals). Instead, it wins on cost per finished task. It resolves a SWE-Bench Pro task with roughly 15,954 output tokens on average, versus 67,020 for Opus 4.8 at max — a 4.2× efficiency advantage. At $6 per million output tokens, that arithmetic adds up fast when you're running agents every day.
Beyond coding, xAI reports Grok 4.5 works for Office automation (Excel models, PowerPoint decks, Word documents) and scored #1 on Harvey's Legal Agent Benchmark for legal document work.
What it means for UK SMBs
Grok 4.5 is free to use in Grok Build, which makes it the cheapest way to experiment with frontier-level agents. The speed (80 tok/s) and token efficiency mean it's well-suited for high-volume, cost-sensitive workflows — think batch content generation, document processing, or automated customer responses where you care about throughput, not peak benchmark scores.
Available now in Grok Build (free) and Cursor (all plans). Not yet available in the EU — expected mid-July.
🧭 The Bigger Picture: What UK SMBs Should Do This Week
Four major releases in one week is a sign that the AI industry is accelerating, not slowing down. Here's a practical action plan:
- Pick one model for your workflow. If you're cost-sensitive, start with GPT-5.6 Luna or Grok 4.5 (free in Grok Build). If you need peak accuracy, Terra or Sol are the new benchmarks.
- Try agent teams. Google's /teamwork-preview is limited to high-tier plans, but the pattern of Planner → Worker → Reviewer → Critic → Auditor is replicable with any platform. Think about which business process would benefit from a multi-step review cycle.
- Go mobile. Codex's phone remote is the most practical productivity unlock this week — the ability to unblock an agent workflow from anywhere is a real time-saver for a small team.
- Price per task, not per token. Grok 4.5's 4.2× efficiency advantage is the metric that matters. When evaluating any AI tool, calculate the cost of a completed task, not the cost of an API call.
Ready to put AI to work in your business?
AI Suite builds practical, UK-focused AI tools that actually finish the job — from invoice chasing to customer support, lead generation to compliance. No hype. Just working systems.
Explore AI Suite Products →📦 AI Suite Products That Help You Act on This Week's News
While the big tech companies release new models, we focus on making AI useful for UK businesses today. Here are the tools that pair with what you just read:
Don't chase models — build systems that finish the job. Explore the full AI Suite playroom →