Research.

Anthropic's engineering team now ships roughly eight times more code per quarter than in 2021–25, with Claude authoring about 80% of it, which drove a 25-fold increase in CI jobs and a 10-fold increase in test volume over six months — a real-world case study in what verification and testing infrastructure has to become once AI-assisted development scales past the pilot stage.

AWS closed new signups for Amazon Q Developer in May 2026 and will discontinue its IDE plugins by April 2027, while Google shut down individual Gemini Code Assist and Gemini CLI plans in June 2026 in favour of its Antigravity platform — leaving Claude Code, Cursor and GitHub Copilot as the three AI coding tools that matter for anyone commissioning AI-assisted development.

GitHub Agentic Workflows, which lets teams describe repository automation in plain Markdown and run it through a coding agent (Claude Code, Copilot CLI, or OpenAI Codex) inside GitHub Actions, has moved from technical preview toward wider availability — search interest in "AI code review" and "agent CI/CD" is shifting from novelty to infrastructure.

Enterprise AI coding startup Factory raised $200m on 15 September 2026 at a $5bn valuation, up from $1.5bn in April, with hundreds of thousands of developers at firms like Nvidia and Adobe already using its autonomous "Droid" agents — a sign that buyers are moving past single AI coding assistants toward platforms that run whole chunks of the development lifecycle.

Anthropic's internal Slack-based coding agent, Claude Tag, now lands 65% of the product engineering team's pull requests, and the company has published case studies on how its product, engineering, and security teams actually use Claude Code in production — a rare look at AI-assisted development at full organisational scale rather than a single-developer demo.

AI product manager postings are running at roughly 714 new US roles a week with no boom-bust cycle, and 47% target manager-level hires versus just 2% junior — a hiring pattern showing that AI-era software teams are being built around ownership and judgement rather than headcount, which changes what founders should expect from a development partner too.

Salesforce and Anthropic announced 'Claudeforce' on 26 August 2026, embedding Claude's reasoning directly into Salesforce's data, workflows and governance — the first time Salesforce has attached its 'force' naming convention, used for Sales Cloud and Agentforce, to another company's product, with the first release, Salesforce in Claude, in open beta from September 2026.

Anthropic's Claude Fable 5.1, released 1 September 2026, more than doubles Fable 5's Terminal-Bench-Science score (24.7% to 52.6%), leads CursorBench 3.2 at 73.4% for agentic coding, and cuts costs by up to 45% on heavily agentic workloads — while adding a zero-data-retention option that removes a real barrier to enterprise AI-assisted development adoption.

Sonar's 2026 State of Code Developer Survey of over 1,100 professional developers finds AI now accounts for 42% of committed code, projected to reach 65% by 2027, while 38% of developers say reviewing AI-generated code takes more effort than reviewing a colleague's — a capacity gap, not just a trust gap, that founders commissioning software should be budgeting for.

OpenAI launched its Agents API in public beta on 10 September 2026, exposing the session orchestration, context compaction and sandboxed execution it built for its Codex coding agent as general-purpose infrastructure any developer can build on, billed through normal model and tool usage rather than a separate fee.

iOS 27, iPadOS 27 and macOS 27 roll out to the public on 14 September 2026, and Apple has already confirmed that from April 2027 every new App Store submission must be built on the iOS 27 / iPadOS 27 SDK or later — a hard deadline for any business planning a new app or a refresh of an existing one.

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 on 1 September 2026, with Fable 5.1 running up to 45% cheaper on heavily agentic coding tasks thanks to a 75% cut on cached input reads, and around 60% fewer false-positive security flags inside Claude Code — a real move in the cost and trust economics of AI-assisted development.

OpenAI told Anysphere on 28 August that it will end Cursor's access to OpenAI models by 12 November 2026, invoking a change-of-control clause after SpaceX completed its $60bn acquisition of the AI coding tool on 14 August — a concrete case study in what happens when a single vendor controls the models behind your development stack.

McKinsey's State of AI 2026 survey of 1,719 organisations found 32% have decided against purchasing at least one software product because agentic coding tools could build the functionality themselves — rising to 41% in tech and nearly half among the highest AI-performing organisations — but the share reporting AI actually moving their bottom line held flat at 37% year-on-year.

GitHub made Claude Fable 5.1 generally available in Copilot on 1 September and GPT-6 Astra generally available on 4 September, joining Gemini models already in the picker — but Claude Fable 5.1 is the first Claude model in Copilot that retains prompts and outputs by default to run Anthropic's safety classifiers, a data-handling detail buyers should know before picking a model.

Salesforce and Anthropic announced 'Claudeforce' on 26 August 2026, an expanded partnership whose first product, Salesforce in Claude, puts live CRM data and 37 prebuilt sales skills directly inside Claude with permissions inherited from existing Salesforce roles; it's in pilot now and moves to open beta in September — a pattern worth understanding if you're commissioning an AI-native product.

OpenAI shipped GPT-6 Astra on 3 September 2026 at 2.5x GPT-5.6 Sol's price after delaying it four weeks over cyber risk, and it's now the first model to hit OpenAI's 'Critical' cybersecurity threshold — scoring 100% on exploit-development benchmarks and finding two previously unknown zero-days during testing — a milestone anyone evaluating AI coding agents needs to understand.

From 30 September 2026, Australia joins the US, UK and EU in Google Play's reduced 20% base commission, and developers meeting quality and integration standards can qualify for the new Apps Experience Program, which cuts the new-install service fee to 15% and drops subscription commission to a flat 10% — a real shift in Android app economics for anyone budgeting a build.

GitHub cut the AI credits included in Copilot Business and Enterprise plans by 36.7% and 44.3% respectively on 1 September 2026, as the promotional top-up from the June move to usage-based billing expired — the subscription price stayed the same, the usage it buys did not.

Anthropic is replacing Claude Code's temporary +50% weekly usage boost with a permanent 25% increase on 14 September 2026 — but because the boost expires the day before, anyone using Claude Code right now sees a 17% cut in weekly capacity, not a rise.

Apple has replaced its per-install Core Technology Fee for EU apps with a flat 5% Core Technology Commission on transactions made through alternative app marketplaces or the web, effective 1 October 2026, as part of a settlement with the European Commission over the Digital Markets Act.

McKinsey's State of AI 2026 survey of 1,719 organisations found that 32% have forgone buying a software product or feature because agentic coding tools let them build it internally instead, with tech (41%) and healthcare (39%) leading the shift — but the survey never asked whether those builds shipped, passed audit, or are still running a year later.

Apple's 9 September event confirmed iOS 27 arrives 14 September 2026 with a refined Liquid Glass interface and a more context-aware Siri, but locked several Apple Intelligence features to iPhone 15 Pro and newer — a compatibility line worth knowing before you scope an iOS release.

Anthropic released Claude Fable 5.1 on 1 September 2026 with a 1M-token context window, 128K max output, and cached-token pricing cut by 75%, alongside Claude Mythos 5.1 — both models are the first from Anthropic to carry invisible watermarking in line with EU AI Act rules.

Salesforce and Anthropic announced Claudeforce, embedding Claude directly into Salesforce's CRM platform in open beta this September, and it's the clearest sign yet that enterprise buyers now expect frontier AI models built into the software they already run, not bolted on as a separate chatbot.

Anthropic's latest Claude Code release adds a fullscreen diff panel, expanded headless and desktop commands, larger inline output limits, and new policy and skill diagnostics — search interest in 'Claude Code update' and 'AI coding agent diff review' is climbing as teams look for ways to review agent-written code faster.

From September 2026, developers must declare whether an app includes social media capabilities before Apple will accept new versions or grant notarization for alternative app marketplaces, triggering a new age-rating content descriptor — a compliance step founders need to budget for before submission, not after.

OpenAI released GPT-6 Astra on 3 September 2026 as a limited preview rolling out to ChatGPT Plus, Pro, Business and Enterprise users plus the API, Azure and AWS Bedrock, with OpenAI President Greg Brockman calling it a plausible marker of the 'AGI era' and the model becoming the first to reach 'Critical' cybersecurity capability under OpenAI's own Preparedness Framework.

Cognition, maker of the autonomous coding agent Devin, is closing a roughly $1bn funding round at a $47bn valuation reported on 2 September 2026, nearly doubling May's $26bn mark as annualised revenue grows from $492m to over $900m in the same period.

Anthropic announced Enterprise Frontier Safeguards on 1 September 2026, replacing its data retention policy after pushback from regulated-industry customers by storing AI misuse-monitoring data in the customer's own cloud infrastructure instead of Anthropic's, rolling out free across Claude Code, Claude Enterprise, Bedrock and Microsoft Foundry from this autumn.

A Broken Object Level Authorisation flaw in Lovable, the $6.6bn vibe-coding platform, let anyone with a free account read another user's source code, database credentials and customer data using five API calls — and stayed exploitable for 48 days after a researcher's HackerOne report was closed without escalation.

OpenAI confirmed on 3 September 2026 that GPT-6 Astra is the first model to cross its internal 'critical' cybersecurity threshold — able to find and exploit previously unknown vulnerabilities across well-protected systems without step-by-step human guidance, including two real zero-days found during testing.

Anthropic's new Enterprise Frontier Safeguards, built with over 100 customers across healthcare, financial services and the public sector, pairs zero data retention with misuse monitoring that runs on infrastructure the customer owns and controls — replacing a data retention policy that had drawn complaints from regulated buyers.

AI agent security and governance startup Zenity raised a $125m Series C on 3 August 2026, led by Norwest with SoftBank Vision Fund 2 and Intel Capital among the backers, after tripling revenue for two straight years — evidence that the risk of letting autonomous AI agents take real actions in production is now a funded category, not an edge case.

NHS England is rolling Microsoft 365 Copilot out to 505,000 clinicians and support staff by October 2026, after a trial across 90 organisations found it saved an average 43 minutes of admin time per person per day — the clearest evidence yet that AI software procurement in the NHS has moved from pilot to default.

Anthropic announced a 'permanent' 25% increase to Claude Code's weekly usage limits from 14 September 2026, then had to admit — after deleting the original post — that it's actually a 17% reduction from today's temporarily boosted allowance, a reminder that AI coding tool capacity isn't a fixed cost.

Anthropic announced Enterprise Frontier Safeguards on 1 September 2026, replacing its prior data retention approach after complaints from financial services, healthcare and public sector customers, storing activity data in infrastructure the customer controls rather than Anthropic's own — a direct response to the compliance concerns that have been the biggest brake on AI coding tool adoption in regulated industries.

OpenAI notified Cursor on 28 August 2026 that it will wind down its contract supplying OpenAI models to the AI coding tool, with a proposed shutoff date of 12 November 2026, following SpaceX's $60bn acquisition of Cursor closing on 14 August — meaning teams standardised on Cursor for GPT-model access lose that route inside eleven weeks unless they bring their own API key.

Search and vendor language is quietly moving away from "no-code" and "low-code" even as the underlying market keeps growing toward a projected $52bn in 2026, because AI-native tools that build software from a plain-language description have made the old drag-and-drop, visual-builder framing feel dated — a naming shift worth knowing if you're commissioning software and hearing pitches in outdated terms.

Google released Gemini 3.8 Flash (codename "Skimaki") on 2 September 2026, a coding- and agent-focused model priced from around $0.75 with a 1M-token context window, positioned to compete directly with Anthropic's Claude Opus 5 and OpenAI's Codex models on software engineering benchmarks — search interest in "Gemini coding" and model comparisons has jumped accordingly.

From 14 September 2026, Anthropic is permanently raising Claude Code's standard weekly usage limits by 25% above the original baseline for Pro, Max, Team and seat-based Enterprise plans — but because that date also ends the temporary 50% summer boost, most active users will see their real weekly capacity fall by around 17%, and searches comparing "Claude Code limits" before and after the change have spiked as teams try to work out what it means for their budget.

OpenAI notified SpaceX on 28 August 2026 that it will terminate Cursor's access to its models on 12 November, citing Elon Musk's history of breaking contracts, weeks after SpaceX closed its $60bn acquisition of Cursor-maker Anysphere on 14 August — a concrete example of the vendor-consolidation risk this signal flagged back in July.

Google launched Gemini 3.8 Flash on 2 September 2026, its third Flash-tier model release in six weeks after 3.6 Flash and 3.7 Flash, at the same $0.75/$3.75 per-million-token introductory price and available same-day inside Google AI Studio, Android Studio and Gemini Enterprise — a release pace that's becoming the norm across every major AI coding model vendor, not just Google.

Anthropic has warned that infostealer malware — including Vidar, LummaC2, StealC, RedLine, and Atomic Stealer on Mac — is lifting active Claude session cookies from infected machines and using them to consume victims' paid usage, bypassing MFA entirely because the stolen session is already authenticated.

Anthropic is replacing Claude Code's temporary 50% weekly usage boost with a permanent 25% increase over its pre-May baseline from 14 September 2026 — a change Anthropic itself confirms works out to roughly a 17% reduction for anyone currently running on the boosted allowance.

Google has released an alpha GenUI SDK for Flutter that lets an AI agent generate and assemble app interfaces at runtime from a developer's own widget catalog, described at Google I/O 2026 as the most architecturally significant Flutter announcement in years.

Anthropic's own usage data from 1.2 million sampled Claude Cowork sessions shows business process and operations work is the single largest category at 33.4%, with software development making up just 8.7% — echoing Epic Systems, where over half of Claude Code usage now comes from non-developer roles.

From 1 October 2026, Apple is replacing the per-install EU Core Technology Fee with a flat 5% Core Technology Commission on digital transactions made through alternative app marketplaces or a developer's own website, alongside a 26% in-app purchase rate, 20% for alternative payment processing, and 15% on web link-outs.

OpenAI reset paid ChatGPT Work and Codex usage limits on 8 August 2026 and then brought back the five-hour cap for Plus accounts on 26 August, the latest swing in a pattern of limit changes that's been running since Codex's five-hour cap was first removed in late July — a reliability signal worth weighing alongside raw capability when picking an AI coding tool.

Nvidia is closing in on a reported $13bn acquisition of Hugging Face and has already agreed a $6bn deal for Poolside, while Stripe bought open-weight model marketplace OpenRouter for over $7bn in mid-August 2026 — three major acquisitions of open-weight AI infrastructure companies inside a few weeks, a consolidation wave that matters for anyone whose AI coding tools or AI products depend on open-weight models underneath.