24 April 2026: GPT-5.5 Arrives and Sierra Buys Its Way Into Agentic Growth (AM)

OpenAI lifted the lid on GPT-5.5 yesterday, billing it as the smartest model it has shipped and positioning the release at the centre of what TechCrunch is calling an emerging AI “super app”. The launch dominates this morning, but Sierra’s acquisition of French startup Fragment, a careful Google push for its in-house AI silicon and a candid Anthropic postmortem round out a busy 24 hours.

OpenAI released GPT-5.5 on Thursday, positioning it as the foundation for a growing product bundle that now looks more like an AI super app than a single chatbot. The company describes the model as its smartest to date, with faster performance across coding, research and data analysis, and the ability to move between tools inside Codex, ChatGPT and the API. TechCrunch reported that the update brings OpenAI “one step closer” to a super app strategy, where one subscription covers writing, research, automations and a fast-growing web of agentic features.

The model shipped alongside a full system card, so developers can read OpenAI’s own safety and capability testing before rolling it into live products. For UK users, availability through ChatGPT Plus and the API is immediate, with Enterprise and Edu tiers following. Pricing was not disclosed in the launch post, though OpenAI said cost per task should fall as efficiency improves. Consumer users on £20 Plus plans should see GPT-5.5 become the default within hours.

The practical test for small businesses is whether GPT-5.5 handles long multi-step tasks without losing the plot, which has been the key failure mode in earlier releases. Watch for independent benchmark runs over the next 24 hours.

Sierra, the AI customer service agent startup co-founded by former Salesforce co-chief Bret Taylor, has acquired YC-backed French startup Fragment. The deal, announced Thursday evening, folds Fragment’s team into Sierra’s agent platform and signals that consolidation in the customer service AI space is running at pace. Sierra has been one of the best-funded agent startups since its 2024 launch, and the Fragment acquisition gives it European engineering talent and additional tooling for agent orchestration.

For UK consumers, the acquisition matters because Sierra already powers customer service flows for several retailers and telcos. More capable agents mean fewer chatbot loops, faster resolution of refunds and shipping queries, and over time fewer calls routed to humans. The trade-off is familiar: better automation typically means lower call-centre headcount. Financial terms of the deal were not disclosed.

Google published a long-form explainer on how its tensor processing units power an ever-growing set of AI workloads, pitched squarely at enterprise buyers weighing up Nvidia alternatives. The post positions TPUs as the backbone of Google’s consumer AI products from Search to Gemini, and makes the case that the custom silicon is now competitive with general-purpose GPUs on price per token for large-scale inference. Google said demand for TPU capacity has forced it to expand its European data centre footprint, with an Austrian site announced the same day expected to create 100 direct jobs.

For UK small businesses, TPUs rarely come up by name, but they shape the economics of any Google Cloud AI service, including Vertex AI, the Gemini API and Workspace AI add-ons. If Google keeps TPU cost curves falling faster than Nvidia’s GPU supply can meet demand, UK Workspace subscribers may see AI features hold their current price point rather than drift upward.

OpenAI also rolled out Automations inside Codex, giving users a way to schedule recurring tasks or trigger them from events rather than running them manually each time. The feature sits inside the Codex academy and lets teams kick off reports, summaries and recurring workflows without writing glue code. Automations can be triggered on a schedule, by webhook, or via API call, and run inside the existing Codex sandbox with familiar permissions.

This is one of the more useful practical updates of the morning. A UK small business using Codex for weekly financial reporting, competitor monitoring or inbox triage can now turn a repeatable prompt into a scheduled job in a few clicks. That removes one of the bigger frictions of agentic work, which has been the need to remember to press “go” on Monday mornings.

Anthropic published an unusually detailed postmortem on recent Claude Code quality complaints, acknowledging that the coding tool had been under-performing for weeks. The engineering writeup walks through the causes of the regression, the telemetry that surfaced it late, and the remediation steps that have now shipped. Anthropic said it has reworked its evaluation harness so future regressions should be caught within hours rather than weeks.

For paying users of Claude Code, which includes a large segment of UK developers and agencies, the post sets a clearer expectation on how quality issues will be handled. It also implicitly concedes what many users had been reporting on forums: that Claude Code genuinely had been worse for a stretch, and the fix is real rather than cosmetic.

Here is everything else worth knowing from this morning’s AI news.

• GPT-5.5 system card published. OpenAI’s full safety and capability writeup went live alongside the model release, useful for developers reviewing the update before production use.
• Google opens first Austrian data centre in Kronstorf. The site will bring 100 direct jobs and add European AI infrastructure capacity, announced the same day as the TPU explainer.
• Noscroll launches an AI doomscrolling replacement. The bot reads the internet for you and surfaces summaries, a neat consumer wellbeing play that may or may not stick.
• Era raises $11M for AI hardware platform. The startup is betting on rings, pendants and glasses as the next wave of AI form factors and wants to sell the software stack that runs them.
• Beehiiv ships webinars and customisable paywalls. Creator-economy tools including AI-assisted workflows, part of a tilt toward becoming the all-in-one creator stack.
• AI galaxy hunters add to the global GPU crunch. Astronomers now compete with model training runs for the same hardware, pushing rental prices higher.
• Tolaria launches an open-source Markdown knowledge base. A useful indie macOS tool for anyone looking to keep an AI-friendly notes system off the cloud.
• Delve customer Context AI discloses security incident. A reminder that AI compliance certification is only as strong as the company doing the auditing.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.