AI News This Week: GPT 6, Claude Haiku 5.5, Microsoft CRM Agents and ChatGPT Ads
What Changed in AI This Week? GPT-6, Claude Haiku 5.5, Microsoft CRM Agents and ChatGPT Ads
The biggest AI developments for UK SMEs and mid-market organisations, from cheaper high-volume models and adaptive interfaces to CRM agents and a new advertising channel.
What if you could ask ChatGPT for a working calculator instead of instructions on how to build one? Or have an AI agent update your CRM, prepare customer responses and handle the repetitive work that usually eats into your day?
What if you could ask ChatGPT for a working calculator instead of instructions on how to build one? Or have an AI agent update your CRM, prepare customer responses and handle the repetitive work that usually eats into your day?
Those are becoming increasingly realistic possibilities, and this week's announcements offer a glimpse of how businesses might actually use them.
OpenAI has introduced interactive tools directly into ChatGPT, Anthropic has made high-volume AI considerably cheaper, and Microsoft is bringing more CRM functionality into the applications employees already use. Meanwhile, ChatGPT advertising is steadily developing into a potential new marketing channel, and Google is using AI to tackle the legacy systems many businesses have struggled to modernise.
There's plenty worth paying attention to, particularly if you're trying to work out where AI could make a practical difference in your organisation, so shall we kick off?
Headlines at a glance
GPT-6 in ChatGPT
Intelligent UI can generate charts, forms, controls and small interactive tools inside a conversation, with rollout underway across paid and free tiers.
OpenAI and Atlassian
GPT-6 models will power agents across Atlassian and Rovo, using the Teamwork Graph to ground work in projects, documents and decisions.
Claude Haiku 5.5
Anthropic says its new small model costs around 75% less to run than Haiku 4.5 and is designed for high-volume tasks such as classification, summaries and live support.
Microsoft Copilot and Dynamics
Thirty new CRM skills bring sales, service and marketing actions into Copilot, while autonomous service and sales agents move further into preview.
ChatGPT Ads
OpenAI is adding visual ads during image generation and expanding attribution, conversion and brand-suitability tooling.
Google Cloud Modernize
A new portfolio uses Gemini-led assessment and migration tools to analyse legacy estates, build cost cases and modernise applications.
GPT-6 turns ChatGPT answers into interactive tools
OpenAI has started rolling out GPT-6 with Intelligent UI across ChatGPT. Instead of returning every answer as a stream of text, ChatGPT can now assemble native interface components such as graphics, buttons, forms, charts and interactive experiences inside the conversation.
That may sound like a cosmetic shift, but it potentially changes what a non-technical employee can make on demand. A finance user could ask for a simple scenario calculator. An operations manager could turn a process into an interactive checklist. A sales team could compare options in a structured interface without first commissioning a spreadsheet or internal app.
OpenAI says GPT-6 can also begin answering while it continues to reason, and that GPT-6 Instant starts responding to web-search questions 44% sooner on average than GPT-5.6 Instant. Those are OpenAI's own evaluations, so they should probably be treated as the average sales and marketing claims rather than concrete proof, but it’s clear that AI is being made to feel faster and less like a blank prompt box.
The rollout began on the 7th of October for Plus, Pro, Business and Enterprise users, followed by Free and Go. Enterprise access remains subject to administrator settings. Importantly, OpenAI says this changes the Chat experience only; the models behind ChatGPT Work and Codex are not changing as part of this release.
Don’t judge Intelligent UI only by whether it looks polished. Test whether a generated tool produces repeatable results, handles edge cases and saves enough time to justify becoming part of a real process. Anything used for pricing, compliance, finance or customer decisions still needs a named owner and a review route.
OpenAI and Atlassian connect GPT-6 to enterprise knowledge
OpenAI and Atlassian also expanded their partnership, bringing models from the GPT-6 family into Atlassian's platform and Rovo. The important layer here is Atlassian's Teamwork Graph, which connects people, projects, documents and decisions so agents can work with organisational context, rather than a single prompt in isolation.
If you’re part of an organisation that’s already running Jira and Confluence, this is one for you. The value of workplace AI increasingly depends on whether it can find the right ticket, understand the current project state, follow the relevant documentation, and leave an auditable update in the system where work is managed.
Atlassian is also using the tools internally. OpenAI says more than 3,000 Atlassian developers use Codex across terminals, development environments and code review workflows. Through Teamwork Graph-powered plugins, those users can access relevant work items and technical documentation while they build and test software.
Connected context can make an agent more useful, but it also makes permissions more consequential. Before enabling broad access, review stale projects, inherited permissions, confidential spaces and the quality of the documentation the agent will treat as truth.
Claude Haiku 5.5 makes high-volume AI materially cheaper
Anthropic launched Claude Haiku 5.5 on the 7th of October as its fastest and cheapest small model, aimed at repetitive, time-sensitive and high-volume work. Anthropic highlights summaries, document compaction, database queries, classification, live customer support and browser use as suitable workloads.
The price change is the headline that caught our eye here, as for prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Anthropic says the model costs around 75% less to run on average than Haiku 4.5. It also halved Sonnet 5.5 cache-read pricing, which - it says - reduces the cost of most agentic work on that model by around 20%.
Haiku 5.5 is also the first Haiku-class model with an adjustable effort setting, meaning model selection doesn’t have to be a binary choice between ‘cheap’ and ‘clever’. A business can decide how much reasoning to buy for a task and reserve larger models for the moments where judgment or complexity warrants the extra cost.
The sensible question here is to ask whether a smaller model can meet a defined quality threshold for a particular workflow. Classification, first-pass triage and retrieval can often tolerate a different model profile from contract review or a customer-facing recommendation.
Route work that goes into AI by risk and complexity. Start with a small model for bounded, repetitive tasks, measure its error rate and escalation rate, then move only the difficult cases to a more expensive model, which is usually a better cost-control strategy than choosing one model for everything.
Microsoft brings CRM skills and agents into Copilot and Teams
Microsoft announced 30 pre-built skills across Dynamics 365 Sales, Customer Service and Customer Insights. The skills bring customer context and actions into Copilot: sellers can surface at-risk deals and propose record updates, service teams can research cases and draft responses, and marketers can prepare outreach with consent and readiness checks.
The skills are generally available in Copilot Cowork, while Autopilot support is in private preview, and Code access is available through Microsoft's Frontier programme. The useful near-term news for you is the availability of guided skills in Cowork; the more autonomous experience is still being tested.
Microsoft also described a private-preview Service Agent that can understand an incoming customer request, retrieve organisational knowledge, draft a reply for approval and create a CRM case when formal tracking is needed. Its Sales Development Agent can research prospects, personalise outreach, manage follow-ups and hand qualified opportunities to sellers, using its own Exchange account and connections to leading CRM platforms.
This is Microsoft's clearest expression yet of what it means by bringing agents into the flow of work - CRM remains the system of record, but employees increasingly interact with it through Teams, Outlook and Copilot. For growing firms, that could remove plenty of copying, chasing and switching between systems. It can also automate a bad process with impressive speed.
Map where an agent can read, write and send before judging the demo. Require review for customer communications and material CRM changes until accuracy, consent handling and escalation behaviour have been tested on your own data.
ChatGPT advertising becomes a more serious acquisition channel
OpenAI is expanding ChatGPT Ads with a visual format that will initially be tested during image generation for Free and Go users in the United States. The ads will be labelled and separate from the generated image, and OpenAI says advertising will not influence ChatGPT's answers.
For businesses, the more consequential update may be measurement, as OpenAI announced integrations with Hightouch, Tealium and LiveRamp for conversion data, plus support from a broad set of attribution and measurement partners. It’s also working with DoubleVerify and Integral Ad Science on brand-suitability pilots designed to assess placement safeguards without giving those partners access to private conversations.
OpenAI shared encouraging early partner figures, including a reported 15.3% lower attributed cost per acquisition for WeightWatchers than its blended paid-search benchmark. These are selected early results supplied by OpenAI and its partners, so it’s not a guarantee of performance for other advertisers. The format is also starting with a limited US test, so those in the UK should treat this as a channel to watch rather than something to budget for immediately.
Prepare for AI discovery without chasing it blindly. Keep product information, pricing and landing pages accurate and machine-readable, and insist on incrementality testing when access widens. A new attribution dashboard isn’t the same thing as proof that a campaign created new demand.
Google applies agents to cloud and application modernisation
Google introduced Google Cloud Modernize, a portfolio that brings its migration and modernisation tools into one place and adds agentic capabilities across infrastructure assessment, platform migration and application modernisation.
The new ‘Modernization Hub’ can analyse source code, map dependencies and produce recommendations for Java, .NET and mainframe applications. In Migration Centre, an Agentic Quick Estimator can convert VMware inventory data into total cost of ownership projections and enable teams to test assumptions via a chat interface. Google says this can reduce weeks of spreadsheet modelling to minutes, although any resulting business case still needs human validation.
Google also put an EKS-to-GKE migration agent into public preview. It automates discovery, manifest translation, storage and network mapping, with human approval gates and in-memory credential handling built into the workflow.
It’s maybe less accessible than a new chatbot feature, at least immediately, but it may matter more to established mid-market businesses. Much of the friction around AI adoption stems from legacy applications, undocumented dependencies, and data that is difficult to access safely; using AI to expose and reduce that technical debt can be a more valuable first step than adding another assistant on top of it.
Treat AI-led assessment as a way to accelerate discovery, not outsource accountability. Validate dependency maps, cost assumptions, licensing implications and rollback plans before an automated migration touches production systems.
Fifosys view
There's a temptation to look at every new AI announcement and wonder whether it's something your business should be rushing out and saying, ‘we HAVE to do this right now!’
But with new models, features and integrations appearing almost every week, keeping up with everything isn't particularly realistic. And adopting tools simply because they're available isn't much of a strategy either.
Let’s take Microsoft's new CRM agents, for example: they could save sales and service teams considerable time, but only if the underlying customer data is accurate and the processes they follow make sense. Similarly, cheaper AI models sound appealing, but the savings won't mean much if they're producing work that needs constant correction.
The question worth asking is where these developments could genuinely help your business, and what needs to be in place before you start using them, which is exactly the kind of discussion we're continuing in our upcoming AI in Business: Part 2 webinar on the 23rd of October.
We'll be looking at how organisations can move beyond experimenting with AI and start putting it to work, including the practical considerations around implementation, integration, and risk management.
This week's AI news FAQs
What is GPT-6 Intelligent UI?
GPT-6 Intelligent UI allows ChatGPT to create interactive elements such as charts, forms, controls and small tools within a conversation. Businesses should still test accuracy, repeatability and suitability before using generated tools in operational processes.
How does the OpenAI and Atlassian partnership affect businesses?
The expanded partnership brings GPT-6 models into Atlassian and Rovo, using the Teamwork Graph to connect work across projects, documents and decisions. Organisations should review access permissions and documentation quality before enabling agents.
What is Claude Haiku 5.5 best suited for?
Claude Haiku 5.5 is designed for high-volume, time-sensitive work such as classification, summarisation, triage and customer support. Anthropic says it costs around 75% less to run on average than Haiku 4.5.
What can Microsoft Copilot CRM skills do?
Microsoft has introduced 30 pre-built skills across Dynamics 365 Sales, Customer Service and Customer Insights. They can help teams retrieve customer context, propose CRM updates, draft responses and prepare outreach, subject to product availability and appropriate controls.
Are ChatGPT Ads available to UK businesses?
The visual ad format described in this roundup is beginning with a limited test for Free and Go users in the United States. UK businesses should monitor availability and assess measurement and incremental results before committing budget.
What is Google Cloud Modernize?
Google Cloud Modernize brings together AI-assisted tools for assessing legacy applications, mapping dependencies, estimating migration costs and supporting cloud modernisation. Businesses still need to validate the findings and approve production changes.
What should SMEs prioritise before deploying AI agents?
Start with a defined business process, check the accuracy of connected data and set clear access permissions. Measure error rates, review time and operating costs, with human approval for consequential customer-facing or system-changing actions.
Turn AI capability into controlled business value
We can help you identify useful AI workflows, understand the systems and data involved, control access and build governance around what happens next.
Talk to our team
Bring us the workflow, AI tool or business process you’re considering and we’ll help you work through the practical next step.
AI in Business: Part 2
Join us on the 23rd of October to explore practical AI implementation, integrations and risk management.