Below I’ve included a straight answer to “what do we do with this?” – with concrete use cases, an implementation path that won’t sink six months, and where Dragon AI could slot in (if you need us to).

What it is (in business terms)

• Agent Builder: a visual canvas where you drag steps (retrieve files, call tools/CRMs, branch on conditions, request user approval), run previews, and version the workflow like software. It’s how your team sees and governs what the agent actually does.
• Connector Registry: one admin place to authorise data sources (Drive/SharePoint/etc.) and tools safely across both ChatGPT and the API. Turn access on/off centrally; stop OAuth chaos.
• ChatKit: a production chat UI you can drop into your product/portal. It already handles threads, streaming, and interaction patterns so your designers aren’t reinventing chat.
• Evals (upgraded): create datasets, grade full traces end-to-end, and auto-optimise prompts against the failures that cost you money. Fewer dark corners, more measurable improvement.

Think of it as moving from “smart chatbot project” to measured workflow automation with safety gates, telemetry, and a UI your customers or staff will want to use.

Where it pays off first (realistic wins in 4–8 weeks)

1. Support triage and resolution
   Route 30–60% of tickets to self-serve, escalate the hairy ones with full context assembled. Guardrails ask for approval before risky actions; trace grading shows exactly where the handoff broke.
2. Sales research and outreach prep
   Let an agent compile firmographics, extract buying triggers from documents/calls, and draft a structured first touch (inside your CRM) with “why this, why now” reasoning you can audit.
3. Ops copilots (SOPs with hands)
   Turn your SOPs into an agent that reads the latest policy, opens the right systems, and walks the user through a compliant workflow—documenting every step for QA.
4. Knowledge copilots for the field
   Onboards and upskills faster by retrieving from controlled repositories, surfacing change notes, and capturing gaps back to the team. Central connectors and versioned workflows mean answers aren’t going rogue.

Why this is safer than previous DIY stacks

• Central governance: The new admin/tenant model and Connector Registry put authorisations in one place rather than scattered per app.
• Transparent behaviour: Visual workflows + versioning make audits and sign-offs sane (legal finally has a thing to read).
• Measured quality: Evals’ trace grading looks at the whole run, not just final text, so you can prove improvements instead of arguing about vibes.
• UI you can own: ChatKit themes and widgets let you ship a consistent, branded surface so no “prototype UI” risk in production.

What about cost and model choice?

OpenAI’s framing is simple: AgentKit pieces are available today (Agent Builder and Connector Registry are in beta), and you pay standard model rates while using them. That means you can keep your current cost controls (max tokens, model mix) and still get the orchestration, governance and eval layers that used to require extra vendors or internal platform teams.

Implementation path that won’t derail the quarter

Week 0–1: Scoping & guardrails

• Pick one job with a clear measurable: e.g., reduce “first response time” by 30% on Tier-1 tickets.
• Identify the minimum viable toolset (file search, web, CRM write) and the hard no-go actions (refunds >£X require approval).
• Set up tenancy + Connector Registry authorisations (Drive/SharePoint), and create a tiny eval dataset (10–20 real cases).

Week 2: First workflow in Agent Builder

• Build the path: intake → retrieve → propose action → (guardrail) → execute or escalate.
• Run in-canvas previews on your eval set; fix the obvious misses before anyone else sees it.

Week 3: Ship the UI

• Embed ChatKit in a private URL with your brand theme; add a couple of widgets for common shortcuts (“Summarise attachment”, “Create Jira”).
• Roll to five friendly users; record traces for every run.

Week 4: Close the loop

• Use trace grading to find the failure patterns; accept auto-prompt suggestions where they help; lock v1.
• Decide whether to add a bespoke tool (e.g., ERP action). If tool selection is flaky, schedule reinforcement fine-tuning (RFT) on o4-mini in a follow-up sprint.

Deliverable at week four: a measured agent with a branded UI, centralised governance, and a scoreboard your COO understands.

KPIs that belong on the dashboard

• Coverage: % of tasks the agent handles end-to-end without human intervention.
• Safety gates: # of blocked risky actions; time-to-approval where required.
• Resolution quality: Eval score on your trace graders; human QA spot-checks.
• Latency & cost per successful task: Token + tool spend divided by resolved cases.
• Change velocity: Time from “workflow tweak needed” to shipped new version.

Common traps (and how this stack helps)

• Shadow connectors: Random OAuth everywhere. → Fix with Connector Registry + tenant policies.
• Prompt drift: No one knows which prompt is live. → Fix with versioned workflows and Evals.
• Prototype UI in prod: Ends up brittle. → Fix with ChatKit and its supported interaction patterns.
• Unmeasured wins: “Feels faster” ≠ board-level impact. → Fix with eval datasets + trace grading from day one.

How Dragon AI can help (light touch to full build)

1) Agent Opportunity Sprint (1 week)
We map 3–5 high-ROI workflows, data/tool constraints, and a KPI model your ops and finance both sign off. Output: blueprint + risk register.

2) Pilot Build (2–3 weeks)
We implement your first Agent Builder workflow, wire Connector Registry, embed ChatKit, seed evals, and train your internal owner to iterate without us.

3) RFT & Scale-Up (optional)
If you need sharper tool-use or domain behaviour, we run reinforcement fine-tuning on your traces with custom graders, then harden monitoring and approvals.

4) Governance & Training
We help legal, data, and IT land a lightweight policy that’s finally workable as well as train the teams who’ll maintain it.

Ready to turn “let’s try AI” into a measurable workflow upgrade?
Let’s talk about your first 30-day agent pilot.
Drop us a line at info@dragonai.uk or book a slot on our site—no slides, just a quick working session with your real docs and KPIs.

What’s new in GLM-4.6

• Bigger working memory: Context expands from 128K to 200K tokens, enabling analysis of large repos, policy packs, or multi-doc briefs in a single pass. Max output is listed at 128K tokens.
• Built for coding: GLM-4.6 targets practical coding inside tools like Claude Code and Cline, with improvements in front-end generation quality and general reliability. Z.AI reports lower average token use vs comparison models in its internal tests, and has open-sourced the prompts and agent traces for verification.
• Reasoning + agents: Better task decomposition, tool invocation and search-augmented workflows, making it easier to orchestrate end-to-end jobs rather than single replies. Inputs/outputs are text-only, which suits most enterprise coding, content and research pipelines.
• Release timing: GLM-4.6 was announced on 30 September 2025; the update specifically calls out the 200K context and positioning as a flagship coding model.

Where it helps in practice

Below are concrete use-cases we’re already seeing demand for. Each can be piloted with a narrow scope, clear metrics and a rollback plan.

1. Repository assistants for engineers
   Code search, refactor suggestions, migration notes, unit test generation and change-impact summaries across large monorepos, enabled by the 200K window. Pair with CI to propose diffs for small, well-scoped tasks.
2. Requirements → UI skeletons
   Feed product briefs, brand guidelines and component libraries to generate clean, logically structured front-end scaffolds that your team then hardens. Treat as lint-plus-starter rather than auto-deploy.
3. Policy and contract analysis
   Load long policies, SoWs and compliance docs in one shot, extract obligations, compare versions and produce red-flag summaries for legal/ops review. The long context minimises chunking artefacts.
4. Slide and document automation
   Z.AI’s broader ecosystem includes a Slide/Poster Agent that turns prompts into structured decks. Combine with GLM-4.6 to draft sales or training materials, then route to humans for brand and accuracy checks.
5. RAG-style research and customer support
   Use tool-calling to fetch from approved sources (KBs, policies, product specs), then return cited answers. Start with internal-only corpora and audit outputs before exposing externally.
6. Global content ops
   Improved translation and stylistic consistency across long-form copy for sites, emails and product help, with localisation nuances retained over large passages.

How GLM-4.6 compares

Z.AI positions GLM-4.6 as on par with leading Sonnet-class models on several public leaderboards, and highlights superior results in a battery of real-world coding tasks run inside Claude Code. Crucially, they’ve published the full CC-Bench trajectories for third-party scrutiny, which is the right direction for enterprise evaluations. Treat these as promising signals to verify in your own stack.

Adoption playbook (what we recommend)

1) Pick one high-leverage workflow
Examples: “summarise PRs and propose tests”, “turn policy packs into Q&A answers”, or “draft slides from research notes”. Keep the scope narrow and measurable.

2) Build a guarded evaluation loop
Define success metrics (accuracy, latency, token cost, human-edit time). Use a small golden set of tasks and compare GLM-4.6 against your current model. Store prompts, inputs, and outputs so you can reproduce and benchmark fairly.

3) Control the context
Exploit the 200K window with curated, non-PII reference packs rather than raw dumps. Add a retrieval layer for freshness, but cap sources to reduce hallucinations.

4) Instrument for cost and safety
Track tokens by feature, add guardrails (policy filters, allow-list tools), and keep humans in the loop for anything user-facing until you have stable quality data.

5) Plan for model plurality
Keep your orchestration layer model-agnostic. GLM-4.6’s pricing via the Coding Plan can be compelling; you still want the option to switch models per task as the market moves.

What Dragon AI can do for you

• Model fit and ROI analysis: We’ll compare GLM-4.6 against your current stack on your data and target tasks, including cost-per-task and edit-time saved.
• Pilot build-outs: Standing up repo assistants, policy Q&A, or deck-generation flows with proper telemetry, rate limits and fallback strategies.
• Safety, governance and observability: Red-team prompts, guardrails, data-handling policies and audit trails aligned to your compliance needs.
• Training and handover: Playbooks for engineers, product and compliance, including prompt patterns and failure-mode checklists.
• Ongoing optimisation: Continuous evals, regression alerts, and token-cost tuning as models and pricing evolve.

If you’d like us to set up a low-risk pilot or benchmark GLM-4.6 in your environment, get in touch and we’ll scope a focused, measurable trial.

Why this matters

Until now, most AI models were great at short bursts of help – think answering questions, generating text, or writing snippets of code. But real business tasks are rarely one-and-done. They involve research, planning, multi-step execution, and often juggling multiple tools at once. That’s where Sonnet 4.5 changes the game.

In a demo, the model built a fully functional chat application – 11,000 lines of code – working autonomously over days. That’s closer to the type of digital teammate companies have been waiting for.

What companies can do with Sonnet 4.5

1. Software Development at Scale
   • Automate boilerplate coding, integrations, and debugging.
   • Use Sonnet as a junior developer that never tires, freeing your team for higher-level design and problem-solving.
2. Research & Analysis
   • Run deep market research that requires pulling information from multiple sources.
   • Produce structured outputs like spreadsheets, reports, or competitor comparisons without constant prompting.
3. Operations & Admin
   • Schedule meetings by cross-checking calendars.
   • Read dashboards, generate summaries, and draft status updates.
   • Keep track of tasks across tools like Slack, Teams, Notion, or your CRM.
4. Customer-Facing Support
   • Build custom agents that can troubleshoot issues, navigate your knowledge base, and even take action in customer systems.
   • Scale up service without scaling headcount.
5. Specialist Domains (Finance, Cybersecurity, Science)
   • Handle long-context, detail-heavy work such as scanning security logs, modelling financial scenarios, or digesting long technical papers.

Why now

Every week, AI providers are rolling out new features but Sonnet 4.5 shows that extended autonomy and reliable tool use are finally here. The building blocks Anthropic is shipping (virtual machines, memory, multi-agent orchestration) give companies the chance to build their own advanced AI systems rather than relying on generic assistants.

The bottom line

Claude Sonnet 4.5 marks a turning point: from AI as a chatbot to AI as a dependable digital co-worker. The companies who start experimenting now will be the ones who gain the biggest efficiency and innovation edge.

At Dragon AI, we can help you integrate Sonnet 4.5 into your workflows – whether that’s coding support, research automation, or building custom agents for your teams.

What’s new

• Turn chats into checkouts: Product discovery and purchase happen directly in ChatGPT conversations.
• Powered by ACP: An open, Apache-licensed protocol from OpenAI and Stripe for agentic commerce—designed to plug into your existing systems, keep payments secure, and let you accept/decline orders.
• Discovery by relevance: Products are ranked on how well they match the user’s query/context, not pay-to-play. It’s free to be discovered; you pay a small fee only on completed purchases (refunded on returns).

How the flow works

1. Shoppers search naturally (“durable carry-on under £250”) and ChatGPT recommends relevant products. Ranking is purely relevance-based.
2. Compare options in one place: ChatGPT lists where the item is available. Merchant ranking factors include availability, price, quality, whether you’re the maker/primary seller, and whether Instant Checkout is enabled. Even without Instant Checkout, shoppers still get a direct link to your site.
3. Buy from you: Users confirm details and tap Pay Merchant; you accept/decline, process payment, and fulfil as normal—you’re the merchant of record end-to-end.
4. Payments: Card, Apple Pay, Google Pay, and Link by Stripe; Plus/Pro users may have payment/shipping prefilled (still editable). Payment data flows securely from buyer to you via trusted providers.
5. Post-purchase: You send confirmation, handle returns/support, and shoppers can view orders in ChatGPT and track on your site. (No auto-opt-in to marketing from these orders.)

Where it’s available

Instant Checkout is live in the United States with US merchants today, with the goal to expand user and merchant geographies next year. Etsy is already supported; Shopify is coming soon.

Two routes to get started

1) Etsy or Shopify merchant

If you sell on Etsy or Shopify, you’re eligible without building a custom integration. Apply to participate and you’ll be onboarded on a rolling basis.

2) Custom integration via ACP

For other stacks, you can integrate by:

• Applying to Instant Checkout.
• Providing a Product Feed that ChatGPT ingests for accurate price/availability. Supported formats: TSV/CSV/XML/JSON. OpenAI accepts updates every 15 minutes.
• Implementing ACP Checkout endpoints and webhooks, returning rich checkout state each step.
• Payments via Delegated Payment: Use a compliant PSP. Stripe’s Shared Payment Token is the first compatible implementation.

Practical implications for growth teams

• New high-intent surface: Capture demand at the exact moment people are deciding what to buy.
• Relevance is king: Like SEO for conversations—feed quality, completeness, freshness, and being the primary seller all matter.
• Faster conversion: Prefilled details for eligible users and native wallet support reduce friction.
• You own the customer: Orders, payments, fulfilment, and support remain on your systems.

Readiness checklist

• Catalogue hygiene: Unique IDs, accurate prices, inventory, rich attributes/media in the Product Feed; plan ≤15-minute refresh cadence.
• Payments: Confirm PSP compatibility with Delegated Payment (e.g., Stripe Shared Payment Token) and define your acceptance rules.
• Order ops: Map ACP order events to your OMS, notifications, and returns flows.
• Measurement: Decide how you’ll attribute and track ChatGPT orders alongside existing channels.

How Dragon AI can help

• Discovery audit & feed build: We structure and validate your Product Feed to maximise relevance and freshness.
• ACP checkout integration: We implement the Agentic Checkout spec, webhooks, and state machine, wired to your current stack.
• Payments enablement: We integrate with Stripe’s Shared Payment Token or your PSP under the Delegated Payment spec.
• Governance & scale: Production readiness, monitoring, entitlement management, and playbooks for roll-outs.

As AI consultants, our job is to remain on the cutting edge of AI advancements and whilst developments are becoming more and more routine, if we step out of the bubble, we’re able to see just how quickly AI is progressing.

With that in mind, we’ve put together a guide on the future of AI in business and how it might impact you.

Predictive Analytics

One of my largest assumptions is that predictive analytics will become easily accessible to all businesses, not just the largest companies.

That means, your business will be able to employ algorithms based on historic data to help ‘predict’ a wide range of critical operational and market factors. This includes:

Supply Chain Disruptions and Inventory Needs: Foreseeing potential bottlenecks in the supply chain, predicting future inventory requirements to avoid stockouts or overstocking.

Customer Behaviour and Demand: Anticipating what products or services customers will want next, when they are likely to purchase, and which customer segments are most likely to churn or respond to specific marketing campaigns.

Sales Trends and Revenue: More accurately forecasting future sales figures, identifying periods of high and low demand, and understanding the potential revenue impact of different strategies or market changes.

Market Fluctuations and Opportunities: Identifying emerging market trends, shifts in consumer preferences, and potential new market opportunities before they become widely apparent.

AI Agents Will Revolutionise Operations and Workflows

It’s no secret that one of the limiting factors of AI is the amount of time it can spend autonomously on any given task. My prediction is that by the end of 2025, autonomous AI agents will be able to go off for hours at a time completing a task with minimal supervision.

This could mean…

Comprehensive Market Research and Analysis: An AI agent could be tasked with conducting in-depth market research, continuously monitoring competitor activities, analysing evolving consumer sentiment across various platforms, identifying emerging trends, and compiling a detailed report with actionable insights—all over several hours without needing human intervention until the final review.

Automated Content Creation and Campaign Management: AI agents could autonomously draft multiple versions of marketing copy, create variations of social media posts, design basic visual assets, schedule content across different channels, monitor campaign performance in real-time, and even make minor adjustments to optimise reach and engagement based on incoming data.

Advanced Data Processing and Reporting: Businesses could delegate tasks like cleaning and processing large datasets, performing complex data analysis, identifying anomalies or key patterns, and generating comprehensive operational or financial reports, with the AI agent working independently for extended periods.

Proactive Customer Relationship Management: An AI agent could manage a significant portion of customer interactions, from initial lead qualification and personalised follow-ups to resolving complex customer service issues by accessing and processing information from various internal systems, all while learning and adapting its approach over hours of operation.

Streamlined Recruitment and Onboarding Processes: AI agents could autonomously screen a large volume of job applications against complex criteria, conduct initial rounds of candidate assessments or interviews via text or voice, schedule follow-up interviews with human recruiters, and even manage the initial stages of employee onboarding by providing information and answering common queries.

Enhanced Software Development and Testing: AI agents could spend hours writing or refactoring code for specific modules based on detailed specifications, performing extensive automated testing routines, identifying and flagging bugs, and even suggesting potential fixes or improvements to development teams.

Sophisticated Supply Chain Monitoring and Optimisation: An AI agent could continuously monitor global supply chain data, track shipments, predict potential delays, identify alternative sourcing options, and even initiate adjustments to logistics plans to mitigate disruptions—all with minimal human oversight for extended durations.

Personalised Learning and Development Pathways: For internal training, AI agents could dedicate hours to curating personalised learning materials for employees based on their roles, skill gaps, and career aspirations, track their progress, and suggest new learning modules or resources.

The key shift would be from AI as a tool requiring constant human prompting for discrete sub-tasks to AI as a more autonomous partner capable of managing significant, multi-step projects over extended periods, thereby dramatically increasing efficiency and freeing up human capital for even more strategic initiatives.

The Workforce Will Undergo a Significant Transformation Towards Human-AI Collaboration

A lot of people in the AI space (including Bill Gates) theorise that AI will be the cause of mass unemployment.

Whilst that may be the case much further down the line (not least because of manufacturing bottlenecks to create the infrastructure to allow this to happen), I do think that the workforce will undergo a significant change within the next 3 years.

My prediction is that we will shift from holding AI’s hand through tasks to a more collaborative relationship.

Examples of which could include:

Strategic Brainstorming and Idea Validation: A marketing team could collaborate with an AI that generates multiple campaign concepts based on broad strategic goals. The human team then refines, critiques, and selects the most promising ideas, with the AI further assisting by instantly modelling potential outcomes, budget implications, or resource needs for the chosen concepts. The AI acts as an tireless idea generator and rapid scenario tester, while humans provide strategic direction and nuanced judgment.

Augmented Scientific Research and Discovery: Scientists could task AI with continuously scanning and synthesising vast amounts of research papers, experimental data, and genomic sequences to identify novel patterns or potential hypotheses. The AI wouldn’t just present raw data but would offer preliminary interpretations or highlight areas warranting human investigation. Human researchers would then design experiments to test these AI-generated leads, working with the AI to analyse the results and iterate.

Co-Piloted Complex Decision Making: Business leaders could use AI “chief of staff” agents. Before a major strategic decision, the AI could independently gather all relevant internal data, analyse market conditions, model various scenarios, and present a succinct brief with several well-reasoned options, including potential risks and benefits. The human leader then engages in a dialogue with the AI, asking clarifying questions and exploring nuances before making the final, human-led decision.

Human-Supervised Autonomous Operations: In manufacturing or logistics, AI systems could autonomously manage entire production lines or warehouse operations for extended periods. Human supervisors would transition from direct control to overseeing the AI’s performance, managing exceptions, and focusing on strategic improvements to the automated system, intervening only when the AI flags a novel problem or requires a decision outside its established parameters.

It’s undeniable that AI is having a growing impact on businesses, both small and large.

As models become more and more intelligent, their capabilities and application within real world businesses increase.

But how does it really impact your business?

We’re going to give you a quick guide as to how we’ve been using AI to help businesses thrive.

Data Analysis

Data is the new gold standard in business when it comes to AI. AI’s pattern recognition capabilities is simply beyond the majority of humans walking our planet today.

Yes, there are still some outliers who would be able to outcompete AI in pattern recognition, but the speed and efficiency of AI means it’s far more cost effective to use, rather than hiring an expert in the 99th percentile.

If you have data in the bank, AI is almost certainly going to have a positive impact on your business.

Research

With increasing token context length, AI is now at a stage where it can independently go off and perform 2 weeks’ worth of research in 10 minutes.

Whether its market research or competitor analysis, you can now save on hefty research consultancy fees which is especially useful for smaller businesses.

Content Marketing

A study by MIT Sloan assessed how people perceive content created by humans, AI and human-AI collaborations. Participants were unaware of the content’s origin during evaluation.

The results showed that AI-generated and AI-edited content were often rated higher in quality than content produced solely by humans or human-edited AI content.

Again, smaller businesses are in pole position to benefit from the speed and cheap cost of content production, creating valuable resources to help them market their business and reach more potential customers.

Automation and Agents

It’s common for solo founders or small business owners to engage in menial office tasks which can restrict the time they spend on growing their business.

We predict that in the latter half of 2025, AI agents will be able to competently complete the job role of a full time admin meaning you won’t have to worry about using precious time resources.