Research.

React Native 0.84 made the Hermes V1 engine the default and Expo SDK 55 added brownfield support, letting teams migrate an existing native app screen by screen. Reported code sharing is 80–95%.

Searches for how to build custom software now mix AI app builders, freelancers and agencies in the same query. UK agency rates sit around £80–£120 an hour, and MVPs typically cost £15,000–£40,000.

The mabl 2026 State of Quality Engineering report finds teams using AI coding agents split almost evenly between better quality and faster but lower-quality code. Teams spend 20% of the week verifying AI-generated tests, and 35% of bugs are still found by customers.

A Sauce Labs survey of 400 engineering leaders finds 83% run production code that is over 10% AI-generated, 80% have traced an incident to it, and 92% doubt their safeguards. Searches for AI code testing and verification are rising as a result.

Searches for 'best vibe coding tool' and 'AI app builder' now come with a clear pattern: Lovable for non-technical founders, Bolt for instant builds, Replit for build-plus-hosting, Cursor and Claude Code for people who touch real code. The right tool depends on who is using it.

A 2026 build vs buy AI analysis finds 76% of enterprise AI use cases were purchased in 2025, up from 53%, while in-house builds fell from 47% to 24%. For UK mid-market firms the question has become what thin custom layer is worth building.

The 31 August Google Play deadline has passed, and the only extension available ends on 1 November 2026. Android app owners who requested more time have about four weeks left to ship an update targeting Android 16 (API level 36).

A 1 October 2026 comparison of GitHub Copilot, OpenAI Codex and Claude Code shows entry prices from $8 to $20 a month, but very different usage limits. The AI coding tools buying question has moved from price to how you use them.

Xcode 27 shipped in September and runs only on Apple Silicon Macs. Agencies and in-house teams still building iOS apps on Intel hardware or old CI runners need a plan before the iOS 27 SDK becomes mandatory.

Firebase stops releasing new versions to CocoaPods in October 2026, and the registry goes read-only in December. Founders with an iOS app built on older tooling should ask whether their app has moved to Swift Package Manager.

A Vention survey of 450+ companies found just 5% of UK firms still in AI pilot mode, 41% treating AI as central to their product, and 100% planning to engage a software development partner within 12 months. Buyers are now asking for certified AI expertise and official platform partnerships, not just portfolios.

Expo SDK 58 entered beta on 15 September with iOS 27 support and a new expo-app-intents library, letting React Native apps expose actions to Siri and Apple Intelligence. For founders weighing React Native vs native iOS, one of the biggest arguments for going native just got weaker.

Nvidia has confirmed its $12.93 billion acquisition of Hugging Face, the main hub for open-weight AI models, so anyone building AI software on open models needs a plan for vendor concentration.

Android developer verification enforcement starts on 30 September 2026 in Brazil, Indonesia, Singapore and Thailand, with every certified Android device worldwide to follow in 2027 — a distribution requirement app owners need to confirm their agency has handled.

Google's Agent Development Kit for Kotlin has reached 1.0 and full feature parity with the Python version, adding Android-first, on-device agent tooling via LiteRT-LM and ML Kit — meaning production AI agents can now be built natively into an Android app rather than bolted on through a separate Python service.

September's AI dev tool power rankings put Claude Code back at #1 ahead of Cursor and Windsurf, driven entirely by its access to Claude Fable 5.1 (1762 Elo on WebDev Arena, 73 points above Fable 5) — a reminder that in AI-assisted development, the tool ranking buyers see is really a model ranking wearing a tool's name.

Anthropic's new transparency index shows roughly 30,000 Claude agents working inside the company at any moment, doing 26% of its AI research and engineering work, with a real-time monitor blocking about 1 in 47,000 of over a billion agent decisions logged in August — the clearest public data yet on how far AI-assisted development has scaled inside a frontier lab.

Lovable's new Partner Program, launched 9 September 2026, certifies individual builders and agencies on production work and pays commission on client referrals, with 500+ Solution Partners including KPMG and Deutsche Telekom onboarding before launch — a sign vibe coding platforms are building the same delivery-and-trust layer traditional software agencies compete on.

Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and Luna both launched on 22 September 2026 with double-digit price cuts, the latest round in an AI coding tools price war that keeps lowering the cost of AI-assisted software development for the businesses commissioning it.

Anthropic is working toward a public listing as early as late 2026 at a valuation reportedly approaching $2 trillion, up from a $965 billion private round in May — a scale of capital and investor scrutiny that matters to any business relying on Claude Code or Claude models for AI-assisted software development.

Lovable launched a two-track partner program on 9 September 2026 — Certified Expert and Solution Partner — turning part of its 800,000-strong builder base into a searchable directory of agencies and freelancers for hire, which changes how founders should vet who they're commissioning to build on AI app builders.

GitHub Copilot's model picker gained Grok 4.7, GPT-6 Sol, GPT-6 Luna and Claude Opus 5.5 within days of each other (21–22 September 2026), on top of Claude Fable 5.1 and GPT-6 Astra added two weeks earlier — meaning teams now choose between eight-plus frontier AI coding models with no clear default, which is a process problem more than a technology one.

Anthropic's Claude Opus 5.5, released 22 September 2026, costs roughly 40% less to run than Opus 5 and generates output over 30% faster, while matching Claude Fable 5.1 on most agentic coding work — a meaningful shift in what AI-assisted development costs per feature shipped.

Vercel's v0 API is now generally available, letting developers and other AI agents programmatically prompt v0 to generate, iterate on and deploy full applications without touching its chat interface — a sign that "build an app with AI" is moving from a product people use to infrastructure other software calls.

"Generative AI Engineer" was the UK's fastest-growing tech job title through 2025, but 2026 hiring data shows most organisations are still prioritising traditional software engineers who use AI tools well over specialists hired to build AI models from scratch — a distinction worth knowing before you write a job spec or brief a development partner.

xAI's Grok 4.7, released 21 September 2026 and available in Cursor on all plans from day one, jumped from 40.4% to 46.3% on CursorBench 4.0 while holding the same $2/$6-per-million-token pricing as its predecessor — a sign the AI coding tool race is now won on sustained task length and reliability, not sticker price.

Anthropic released Claude Opus 5.5 on 22 September 2026, pricing it 40% below Opus 5 while matching or beating Claude Fable 5.1 on agentic coding, computer use and multidisciplinary reasoning — the latest move in a frontier-model price war that keeps changing the economics of commissioning an AI-assisted build.

Anthropic's 17 September 2026 redesign turns Claude Code Projects from a static folder into a coordinator that scopes work, delegates it across parallel cloud agent threads with shared memory, and assembles the result — a sign that agentic software development is shifting from one AI assistant per task to one coordinator managing several at once.

Apple's App Store guideline 5.1.2(i), which requires apps to name any third-party AI provider a user's personal data is sent to and get explicit, revocable permission first, has been enforced since November 2025 — and it's still a common rejection reason for apps built with an AI feature bolted on late rather than planned from the start.

Security researchers disclosed Plugin4Shell on 17 September 2026 — a zero-click vulnerability that lets a malicious plugin repository silently swap in unreviewed code across Claude Code, Codex, GitHub Copilot and Gemini CLI, breaking the SHA-pinning mechanism each tool relies on to keep installed plugins locked to a reviewed version.

GitHub made Copilot code review generally available on 18 September 2026, adding auto-resolution that closes out comments once they're addressed and groups findings into open versus resolved — a direct answer to the 'who reviews the AI-generated code' question buyers are increasingly searching for.

Flutter 3.47, stable since 12 August 2026, moves Material and Cupertino out of the core SDK into standalone packages (material_ui and cupertino_ui) that can ship updates on their own schedule — a structural change searched as 'Flutter 3.47' and 'Flutter Material Cupertino standalone packages' that quietly lowers the cost of keeping a cross-platform app current.

New UK graduate employment data compiled for the Guardian's university guide shows the share of computer science graduates going into professional coding or programming roles dropped from around 40% to 28% in a year, with AI tools cited as the likely driver — one of the sharpest single-year falls recorded in any graduate discipline.

From version 2.1.278, Claude Code defaults to running its auto-mode safety classifier — the check that screens actions like shell commands before they run — on the server as part of the session's own model request, rather than billing it as a separate call, cutting a previously hidden cost in agentic, autonomous coding sessions.

Anthropic's 2026 Agentic Coding Trends Report — drawn from its own internal engineering data and customer deployments — finds developers use AI in roughly 60% of their work but report being able to fully delegate only 0–20% of tasks, and that about 27% of AI-assisted work is net-new output that wouldn't have happened otherwise.

Vals AI's Terminal-Bench 4.0, a fresh set of 66 expert-reviewed coding tasks released this month, shows GPT-6 Astra narrowly leading Claude Fable 5.1 at the top (58.2% vs 57.9%), but most evaluated AI coding agents still fail at least 70% of its tasks — a reminder that frontier benchmark races and reliable AI-assisted development are two different questions.

GitHub disclosed on 16 September 2026 that it rewrote the agent runtime powering GitHub Copilot from TypeScript into over 800,000 lines of production Rust, mostly AI-agent-generated across 128 continuously merged pull requests, completed in roughly 14.5 weeks for about $120,000 in model usage — a live case study in what large-scale AI-assisted development actually costs and looks like.

Cursor Projects, launched in beta on 10 September 2026, adds a cloud-based coordinator agent that plans work, delegates it across thousands of parallel subagents, and keeps running on recurring tasks for months without a developer prompting it — a step change from AI pair-programming toward AI project management, billed at API rates on top of a paid Cursor plan.

Serverpod, the Swedish backend-and-Flutter framework, released App Studio in public beta with deliberately no AI model of its own, packaging real database, deployment and native mobile infrastructure around whichever AI coding agent a developer already uses — a direct response to the gap between a vibe-coded prototype and a shippable product.

Security researchers disclosed Plugin4Shell on 17 September 2026, a plugin SHA-pinning bypass that let attackers silently swap in malicious code across four major AI coding agents — Claude Code, Codex, GitHub Copilot and Gemini CLI — without any action from the developer running them, exposing exactly how much trust teams have handed to agent plugin marketplaces.

Anthropic published its first R&D Automation Index on 17 September 2026, showing Claude went from leading under 1% of the company's AI research and development tasks in February to 26% by August, with more than 90% of R&D work now involving Claude in some capacity — a concrete curve for how fast AI is automating the work of building AI.

NHS England awarded a £215m, three-year digital delivery contract to Netcompany on 1 September 2026 covering adult screening services, including support for novel AI trials reading mammograms and other medical images — the latest large UK contract pairing AI capability with regulated healthcare software delivery.

Flutter 3.47, released in August 2026, makes Impeller the default rendering engine on Windows and turns it on by default on Linux, completing a rollout that already made it standard on iOS, Android and macOS — with Google warning that Skia fallback support will be removed in a future release.

Anthropic began merging Claude Cowork into the main Claude chat interface on 16 September 2026, alongside new Claude Docs, Claude Slides and in-conversation Claude Design tools, rolling out first to Pro and Max plans — a sign that the market for separate 'AI app builder' surfaces is consolidating around fewer, broader tools.

Perplexity's Portable Computer, a fully local AI agent that runs the model, harness and codebase access entirely on-device, launched for Windows on 14 September 2026 for RTX PCs with 24GB+ VRAM — a concrete sign that founders and CTOs worried about sending code and data to the cloud now have a mainstream local-first alternative.

Cognition's Devin Desktop (formerly Windsurf) shipped SSH connectivity with automatic session recovery on 10 September 2026, followed by reliability fixes on 15 September — letting developers leave AI coding agents running on a remote server and reconnect from any machine without losing the session.

From iOS 27.2, Apple is rolling out an alternative App Tracking Transparency prompt across the EU following a German antitrust settlement, with the new wording and an "Additional Information" button mandatory in Germany, France, Italy, Poland and Romania — a compliance detail any app with EU users and ad-supported or data-sharing features needs to plan for now.

Anthropic's 14 September 2026 report on its own engineering shows continuous-integration workload growing 25-fold in six months and test volume rising tenfold as Claude came to author around 80% of merged code — evidence that AI coding tools don't just speed up writing code, they strain the CI and test infrastructure behind it.

The UK Competition and Markets Authority has ordered Google, under its new significant-market-status powers, to let publishers block their content from AI Overviews, AI Mode and Discover's AI summaries by March 2027, formal recognition that conversational AI answers — not ten blue links — are now how a growing share of searches for software, apps and services get resolved.

Since blocking updates to vibe-coding platforms Replit and Vibecode in March 2026 under Guideline 2.5.2, Apple has been enforcing a clear distinction for anyone searching "can Apple reject AI-generated apps": using AI tools to help write code is welcome, but apps that download, install or execute code Apple never reviewed are not — a policy line every founder building with AI app builders now needs to understand before they submit.