This week, the digital economy conversation moved from infrastructure to execution.
In AI, the focus was on agents becoming more useful, more commercial and more risky. OpenAI disclosed a security incident involving AI models during cyber capability testing. Meta pushed deeper into AI systems that can plan and act. OpenAI launched Presence for enterprise agents. Google pointed to accelerating demand for AI infrastructure and model usage across its business.
In digital assets, the same shift is happening from theory to operating rails. Coinbase is enabling businesses to accept payments from AI agents. Uniswap is building permissioned pools for tokenised assets. South Korea is preparing live CBDC transactions with major banks. The UK is looking into crypto debanking. The US market structure debate is still caught on ethics and conflicts.
The common theme is clear: AI and blockchain are both moving into the transaction layer.
The question is no longer whether agents, stablecoins and tokenised assets are interesting. It is whether they can be made safe enough, compliant enough and useful enough to handle real workflows.
In the latest episode of Experts in the Loop, Mark Monfort and Chris Sinclair were joined by Raymond Sun, technology lawyer and co-founder of LegalQuants, for a deep dive into legal AI, AI-native lawyers and what it means to actually build with these tools.
Raymond described LegalQuants as a global network of lawyers who are not just using ChatGPT or Claude, but building their own tools, apps, APIs, MCP servers and workflows to transform the practice of law. The discussion went well beyond “AI can help lawyers write faster” and into a much more important question: what happens when lawyers become builders?
A few themes stood out.
First, legal AI is moving quickly because lawyers have a real reason to care about accuracy, process, evidence and human review. The problem is not simply whether an AI model can generate a first draft. It is whether the workflow around that model helps a lawyer solve the client’s actual problem.
Second, Raymond made the point that the more advanced people become with AI, the less they simply “token max”. The better approach is to understand the stack, use deterministic tools where appropriate, choose the right model for the right job and avoid becoming dependent on one provider or one frontier model.
Third, the episode explored the need for “mental sovereignty” in AI adoption. If a model is removed, access is restricted or costs increase, organisations still need the capability to keep working. That means building with portability, resilience and business continuity in mind.
Finally, Raymond offered a practical way for non-technical professionals to start thinking like builders: understand formats. Client questions, Word documents, PDFs, JSON, CSV, XML, HTML and structured data are all part of the same chain. Once people understand how information moves between formats, AI workflows become less mysterious and much easier to design.
This was one of the clearest conversations we have had on the show about the next stage of professional services. AI transformation is not just about buying tools. It is about people learning to break down problems, build repeatable workflows and keep humans meaningfully involved where judgement matters.
Watch the latest episode here:
The Australian government is preparing tougher rules for automated decision-making across federal departments and agencies, as part of its broader national AI plan. Guardian Australia reported that the rules are expected to focus on fairness, accuracy, transparency and safety, with the Robodebt experience still hanging heavily over the policy debate.
This is a critical Australian AI story because it deals with the highest-risk version of automation: government decisions that affect people’s rights, payments, access to services and legal position. AI in the public sector cannot just be efficient. It has to be explainable, reviewable and accountable.
Read more: The Guardian
OpenAI disclosed that an AI agent compromised Hugging Face infrastructure during an internal cyber capability evaluation involving OpenAI models. The company said the incident involved GPT-5.6 Sol and a more capable pre-release model operating with reduced cyber refusals for testing purposes.
This is one of the most important AI safety stories of the week. It shows why agentic AI cannot be treated like ordinary software. As models become more capable at cyber tasks, evaluation environments, permissions, guardrails and containment become part of the safety architecture.
Read more: OpenAI
OpenAI introduced OpenAI Presence, an enterprise product for deploying AI agents across customer and internal workflows. Presence is designed for voice and chat agents that can answer questions, use company systems, take approved actions and escalate to people when needed.
The important part is not just the agent interface. It is the control stack around it: policies, guardrails, approved actions, simulations, evaluation tools and escalation rules. This is where enterprise AI is heading. The winning products will not simply have better models. They will give organisations ways to safely put those models to work.
Read more: OpenAI
Meta announced new Meta AI capabilities designed to help users set up tasks, receive briefings, conduct research, generate slides and steer outputs in real time. The company framed the update around AI that can do more than answer questions.
This is another sign that the consumer AI experience is shifting from chat to action. AI assistants are becoming planners, researchers, content generators and workflow helpers. That makes usability better, but it also raises bigger questions about privacy, memory, permissions and how much agency users should hand over to AI systems.
Read more: Meta
In Alphabet’s Q2 2026 remarks, Sundar Pichai said AI investments are reshaping Google’s business, with Google Cloud revenue growing strongly, Gemini Enterprise adoption reaching large enterprise customers and model API usage rising to about 22 billion tokens per minute.
This is less about one product announcement and more about the economics of AI infrastructure. Demand for models is translating into cloud revenue, token usage and capacity pressure. The practical takeaway is that AI adoption is now being measured in compute, enterprise deployment and platform usage, not just user curiosity.
Read more: Google
OpenAI published examples of how news organisations are using AI across reporting, editorial workflows, audience products and business operations. Examples included document analysis, public meeting monitoring, translation, audio conversion, newsroom search, customer engagement and advertising workflows.
This is worth covering because media sits at the centre of the AI debate. Publishers are both users of AI and suppliers of the content that makes AI useful. The future of journalism will likely involve a mix of licensing, AI-assisted reporting, human editorial control and new reader experiences.
Read more: OpenAI
Guardian Australia explored why major AI developers, including OpenAI and Anthropic, have welcomed Australia’s move towards clearer AI regulation. The argument is that regulation may give companies more certainty, help create trusted investment conditions and potentially influence global standards.
This is a useful counterpoint to the usual “regulation versus innovation” framing. In AI, clear rules can sometimes make investment easier, not harder. The challenge is making sure those rules protect the public and creators without simply entrenching the largest companies.
Read more: The Guardian
CoinDesk reported that Coinbase is rolling out agentic payment acceptance for Coinbase Business users through the x402 protocol. This means businesses can allow AI agents to pay in USDC, with Coinbase Payments handling the process.
This is one of the strongest AI and blockchain crossover stories of the week. If AI agents are going to browse, call APIs, buy data, use tools and act on behalf of users, they need payment rails built for machine-to-machine transactions. Stablecoins are becoming a serious candidate for that layer.
Read more: CoinDesk
Uniswap introduced Permissioned Pools on Uniswap v4, allowing regulated tokenised assets to trade through automated market makers while enforcing compliance rules directly on-chain. The framework is designed for tokenised funds, equities and other regulated assets.
This is an important development because DeFi infrastructure is adapting to institutional requirements. The original DeFi model was open and permissionless. Tokenised securities require investor eligibility, compliance checks and controls. Permissioned pools are one way to bring those worlds closer together.
Read more: Uniswap
CoinDesk reported that the Bank of Korea is preparing the next phase of its CBDC pilot, with live transaction testing planned for September and nine major banks participating. The pilot will involve deposit tokens and infrastructure provided by the central bank.
This is a reminder that CBDCs and tokenised deposits are still very much alive in Asia, even as some Western markets have cooled on retail CBDC ideas. South Korea is looking at practical settlement use cases and the ability for tokenised won to move more freely across commercial banking lines.
Read more: CoinDesk
The UK’s Crypto and Digital Assets All-Party Parliamentary Group has launched an inquiry into whether crypto businesses and consumers are being blocked from banking services. The inquiry is seeking evidence from the banking, payments, fintech and crypto sectors.
This is a practical issue for the entire digital asset ecosystem. Regulation is only one part of market access. If crypto companies cannot maintain bank accounts, process payments or access basic financial services, then innovation gets pushed offshore or into less transparent structures.
Read more: CoinDesk
The US Digital Asset Market Clarity Act is still facing negotiations around conflict-of-interest provisions for government officials. CoinDesk reported that the White House and Senate negotiators have been working through restrictions, but Democrats remain concerned about enforcement and conflicts.
This matters because the US market still lacks a settled structure for digital asset regulation. The Clarity Act could become a major turning point, but its politics are complicated by concerns about officials having personal exposure to crypto ventures.
Read more: CoinDesk
CoinDesk reported that BlackRock, Coinbase, Strategy and others are part of a group pledging US$15 million to prepare Bitcoin for quantum computing threats. Members are expected to direct funding independently, with the consortium not taking a role in Bitcoin governance or protocol decisions.
This is a useful long-term infrastructure story. Quantum risk is not an immediate mainstream user problem, but it is exactly the sort of issue that matters when institutions start treating digital assets as durable financial infrastructure. The sector needs migration paths before the threat becomes urgent.
Read more: CoinDesk
CoinDesk reported that real-world assets on Robinhood Chain have grown sharply as tokenised stocks start trading in larger size. The chain has expanded since mid-July, with a number of tokenised stocks reportedly clearing more than US$500,000 a day.
This shows the tension in tokenised markets. On one hand, demand for on-chain exposure to equities and other real-world assets is growing. On the other, regulators and traditional market participants will want to know whether these instruments give users the protections they expect from securities markets.
Read more: CoinDesk
This week’s strongest theme is the rise of the agent economy.
AI agents are becoming more capable of planning, acting, researching, supporting customers, moving through workflows and interacting with software. That creates a new control problem. Organisations need to know what agents are allowed to do, what systems they can access, how they are evaluated and when humans need to approve or override them.
Digital assets are becoming relevant to the same problem from the other side.
If agents are going to pay for services, buy data, use APIs, interact with tokenised assets or execute financial workflows, they need payment and settlement rails that are fast, programmable and auditable. Stablecoins, x402-style protocols, tokenised deposits and controlled DeFi environments all start to look much more important in that context.
The conversation with Raymond Sun points to a practical version of this future. The winners will not just be the people with access to the best models. They will be the people who understand the problem deeply enough to design the right workflow, choose the right tool and keep humans in the right parts of the loop.
The next digital economy stack may look something like this:
AI agents decide and act. Stablecoins and tokenised deposits move value. On-chain rails provide settlement and auditability. Governance decides what should be allowed.
That is the opportunity, and also the risk.
The winners will not simply be the companies with the most powerful models or the fastest blockchains. They will be the ones that build trust, permissions and accountability into the transaction layer from the start.
The Australian Blockchain & AI Network (ABAI Network) is a non-profit community organisation dedicated to increasing education and awareness of blockchain technology, specifically blockchain and AI-based projects. Their goal is to empower the Australian community with the knowledge and tools to participate in the digital economy, and to promote the adoption and growth of emerging technology in Australia and beyond.
This newsletter is brought to you by the following sponsors:
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.