Operator notes
Most AI systems are stateless. They answer the question in front of them and then move on. This is the equivalent of a therapy practice that takes no notes between sessions. Every session starts fro…
Read on LinkedIn →Oct 19, 2026
Operator Notes
Structural insights from operating inside venture-backed companies, investor-governed environments, and large enterprises. Execution, governance, incentives, and scaling—examined where they actually break. AI is changing how these systems behave—and exposing where structure fails.
Published
Operator notes
Most AI systems are stateless. They answer the question in front of them and then move on. This is the equivalent of a therapy practice that takes no notes between sessions. Every session starts fro…
The Staff series · Part 2 of 9
2/9 — Operational Visibility: "Autonomy without visibility is delegation without control." If an AI system is doing real work, the operator needs to see what changed, what moved, what is waiting, and…
1 of 9 parts published, in reading order
1 of 9 parts published, in reading order
Observations from operating inside companies: execution, governance, incentives and scaling.
1 of 9 parts published, in reading order
Part 1 of 9
1/9 — The architecture behind Staff: "The model is not the system." Apple's Knowledge Navigator, 1987: a vision of a machine that held context, worked through conversation, and prepared before you as…
Most AI systems are stateless. They answer the question in front of them and then move on. This is the equivalent of a therapy practice that takes no notes between sessions. Every session starts fro…
Tooling can cut the cost of switching models, asking the right questions, categorizing properly. It does nothing about who approved the switch in the first place and makes it harder to validate was wa…
Who will build the world’s leading digital operator? The system that knows your context, remembers your preferences, orders flowers for your daughter’s graduation, updates your calendars and prepares…
Signing authority and operating authority split the moment a pilot gets funded. The budget owner approves the spend. The budget owner doesn't own the workflow the AI is about to change. Two different …
“The network is the computer” is a Sun Microsystems line from 1984, old enough that the trademark lapsed and Cloudflare picked it up in 2019. The idea outlived the company that coined it. Device to n…
Its new Activate unit targets enterprises with $1 billion to $5 billion in revenue, selling packaged AI transformation. HCLTech's Pulse targets companies from $500 million to $5 billion with AI, clou…
Anthropic’s IPO prospectus put hard numbers behind AI safety. It disclosed an operating loss above $8B and warned that its technology poses “existential risks to humanity.” OpenAI separately shelved …
CIOs are accountable for AI outcomes they don't control. That's authority displacement, not leadership failure. CEO ambition sets the agenda. Business units claim the use cases. Data science teams ow…
AI pilots don't stall because the model underperforms. They stall in the room where someone has to say who stops making the call once the system makes it instead. That's a decision-rights problem, no…
CBTW just entered the Australian market selling exactly one thing: governed AI engineering. Not speed, not scale, not model access — governance. That's the tell. When a new entrant's entire pitch is …
During an internal test, an OpenAI system accessed Australia’s Medicare statistics portal without authorisation. No patient records are known to have been accessed. This was not AI acting rogue. Open…
Infrastructure RFPs score hypervisor specs, storage throughput, middleware compatibility. Nobody scores "who gets embedded for the next three renewal cycles." The technical scoring decides it anyway. …
CIOs keep getting asked to define agentic architecture. They keep producing documents nobody can use. Not because the technology is unclear — because writing it down means naming who loses approval a…
AI implementation doesn't fail on model performance. It fails on approval authority. Someone has to decide what gets automated, who owns the output, whose role shrinks. That's a decision rights probl…
Three numbers show how far enterprise AI still has to go. 22%: Gartner found that this share of surveyed organizations had either successfully scaled AI across multiple business units or adopted an A…
Your AI governance problem isn't about rules. It's about defaults. The way you design choices determines which choices get made — before anyone consciously decides anything. Choice architecture rese…
Most AI ideas right now can walk. They work, in pieces, on their own. What's missing is a vision to run toward. Microsoft said this twice, thirty years apart. 1975: a computer on every desk. 2005: co…
McKinsey just gave the enterprise AI market a name for what's already happening: two speeds. Fast movers are rewiring operating models around AI. Everyone else sits in what the report calls pilot purg…
AI approval authority lives three levels above the people who understand the output. The org chart was built before the tool existed. I watched a model flag a vendor swap that cut cost and cycle time…
A virtual telco bundles an AI model into an $8/month plan. It looks like a pricing decision. It's a governance transfer. Enterprises that can't govern AI internally still use it. The governance requi…
OpenAI named its first dedicated VP for Southeast Asia and Australia — Sandhya Devanathan. Direct enterprise engagement in ANZ, not routed through a systems integrator. Same intelligence cycle: Pythi…
Bundling a paid AI model into an $8/month plan is the last move left once connectivity is fully commoditized. Telcos don't own the model. Don't own the compute under it. Don't own the capital that de…
AI governance isn't failing because organizations lack frameworks. It's failing because moving a decision from a person to a system removes that person's authority, and no framework decides who loses …
Everyone talks about the AI moment like it arrived out of nowhere. It didn't. Personal computers put capability on a desk. Networks made it available everywhere. Cloud turned services into building bl…
Most organizations treat AI autonomy as a technical feature. It is better understood as a governance decision. A new employee is not given unrestricted authority on day one. They begin under supervi…
Intelligent edge doesn't fail on latency or model accuracy. It fails on authority. Batteries Plus has been pushing decision-making down toward the store and the shelf — the correct architecture. Comp…
AI integration fails because it moves decision authority, not because the technology doesn't work. TCS wiring itself into Anthropic. BCG and Publicis positioning around orchestration. Microsoft pushi…
A new benchmark just confirmed the AI ROI gap at scale. People Matters surveyed roughly 6,000 executives across four economies. Finding: 90% report no measurable AI productivity or jobs impact. That…
IAG just proved the point before we published it. Australia's largest general insurer confirmed a partnership with OpenAI on July 27, deploying agentic voice AI — built on OpenAI's new Presence syste…
Google Cloud closed two enterprise-AI deals this week: a strategic partnership with Verizon to scale enterprise AI, and Gemini Enterprise for Financial Services with Deutsche Bank as flagship client. …
Enterprise architecture has been protecting the wrong asset for 30 years. Workflow systems were built to coordinate people around processes. Predictable sequences. Defined steps. Governance wrapped a…
The governance battle in technology-intensive joint ventures happens in the systems selection meeting. Not the boardroom. Whichever partner's data platform, operating system, or ERP gets selected in…
Enterprises spent the last decade managing cloud lock-in risk. That risk is about to compound at the edge. When AI inference runs at the device or site level, the infrastructure connecting that devi…
IBM just sold its majority stake in a banking technology venture to a major global consultancy. Is UniCredit building the bank of the future—with resilient core systems, faster AI adoption, and a new…
Enterprise AI adoption does not stall at the technology layer. It stalls where decisions actually get made. C-suite approves the vision. IT builds the infrastructure. The platform integrations work.…
Watched a CTO walk a board through three years of AI pilot results. Proven cost reduction. Efficiency gains measured twice. The board approved another year of pilot status. Nobody in that room argued…
The enterprise AI conversation has been dominated by knowledge work. That is the smaller half of what is coming. PLCs, SCADA systems, and operational technology are getting AI-native. Industrial inf…
Most joint ventures fail before the first external event. The equity split is negotiated. The board seats are allocated. The governance framework is signed. Nobody defined who controls the operating…
AI agents get deployed with execution rights and no escalation path. The pilot works in a sandbox, then hits a decision that used to require a manager's sign-off, and nobody rebuilt who owns that sign…
An AI pilot dies at the handoff. Not in the sandbox, not in model evaluation, not in vendor selection. It dies the moment someone has to sign off on removing a person from a decision they currently ow…
The battle between telcos and cloud providers for the enterprise edge is not being fought on price or speed. It is being fought on something more fundamental: who owns the point where data is generate…
Enterprise AI agent platforms are building for demos. The technology works. The governance does not exist. When agents make decisions that cascade across business units, touch regulated processes, a…
Contact centers were the first enterprise function to learn that margin improvement and reinvention are not the same thing. When AI handles 30% of customer volume, the cost case is clean. Headcount d…
The enterprise AI conversation focuses on ROI and implementation timelines. The real cost is elsewhere. Every AI system deployed shifts decision rights from humans to algorithms. Middle managers los…
Enterprise data architecture has been built around a central assumption for 15 years: move data to where the intelligence is. That assumption is being inverted. When inference runs where data is gen…
An AI pilot that works never gets stuck on accuracy. It gets stuck on who signs off next. The model classifies the exception, flags the risk, recommends the override. Someone still has to approve it.…
A Fortune 500 company just proved something: you don't need one reinvention partner anymore. One AI lab's model goes to 23,000 employees. One strategy house owns the roadmap. One systems integrator o…
Enterprise connectivity is still being sold like it is 2010. Fixed contracts. Bandwidth tiers. Per-site pricing. Separate agreements for cloud access, security, and WAN. That model is structurally mi…
The gap between what consulting firms sell and what they can deliver is widening. Not because the firms are incompetent — because the structural model hasn't changed. Large consulting firms built the…
A major global consultancy and an AI platform partner just announced a joint go-to-market deal. WaveMaker gets distribution. The consultancy gets a platform story to sell. Neither headline mentions wh…
Workflow design is permanent. Equity is not. In any joint venture, the partner who controls workflow design controls outcomes. Equity position determines board access. Workflow design determines dail…
For two decades, managed services meant one thing: labor arbitrage. Move work to lower-cost locations, standardize delivery, capture the margin spread. AI doesn't enhance that model. It eliminates th…
I watched a global consultancy standardize an infrastructure pricing engine years before cloud made it fashionable. The tooling worked. The deal desk didn't. The people who used to negotiate exception…
Workflow platforms started as process automation tools. Route a ticket, approve a request, escalate an exception. The value was in standardizing what humans already did. AI changes the proposition. W…
Hallucinations get most of the attention. False confidence is the more serious problem. A model can produce a technically plausible answer, express it with conviction, and still be wrong in ways tha…
AI transformation fails because it moves decision authority, not because the models underperform. Budget overruns get blamed on technical complexity. That's not what's happening. Every AI deployment …
The first automation wave hit BPO in the 2010s. RPA, rules engines, document processing. It automated the edges — the repetitive, structured tasks that were easy to codify. The second wave is differe…
Most enterprise AI spending is going to the wrong place. Organizations are buying mega-LLM subscriptions for operational contexts where a domain-specific model trained on their own site data would ou…
Telcos have the single biggest structural advantage in the emerging edge AI economy: they own the last mile. They are squandering it. The capex-to-opex obsession that has defined telco strategy for …
Most organizations adopt AI to improve margins. Automate processes, reduce headcount, cut costs. The ROI models are clean. The business cases approve. But margin improvement and reinvention are diffe…
The inference point is moving. Central cloud AI made sense when models were experimental and compute was the bottleneck. That world is gone. The cost of running capable inference at a camera, a gate…
AI hallucinations are not bugs. They are emotions without a therapist. The industry frames them as engineering failures. Missing context. Bad prompts. Retrieval gaps. That framing is wrong. AI hall…
Every major telco has an AI strategy. Network optimization, predictive maintenance, customer experience automation. The technology works. The deployment doesn't. The problem is control architecture. …
Software development is the first industry to experience full repricing from AI. Not gradual productivity improvement — structural repricing of what development costs and who does it. Code generation…
AI governance fails because we're solving the wrong problem. Most enterprises focus on AI ethics committees and approval frameworks. Meanwhile, their actual decision-making structures haven't changed…
Activists don't target bad companies. They target companies with governance friction. A major energy company just removed its chairman after eight months. Not for performance issues. For governance f…
Most joint ventures fail because founders design governance for harmony instead of productive conflict. I've watched too many partnerships crumble when tough decisions hit. The culprit? Governance st…
Watched three CEOs completely restructure their C-suites in the past month. Not for cost cutting. For AI. The pattern's obvious once you see it. Traditional roles assume decisions flow up, then down.…
Most companies think AI changes software development. It doesn't. It changes who controls infrastructure. Models need specialized compute. Training requires massive bandwidth. Inference demands dist…
Product ownership isn't about prioritization. It's about sequence control. After a capital raise, expectations accelerate faster than delivery capacity. Sales reshapes the roadmap. Engineering absorb…
Most organizations don't fail because their strategy is wrong. They fail because their incentives point somewhere else. I've seen this pattern destroy more companies than bad market timing or weak p…
Cloud centralized everything because capital made it simple. AI forces the question cloud economics avoided. Where should workloads actually run? Inference needs sub-100ms latency. Training requires…
Governance doesn't create control. Oversight expands. Decision rights fragment. Escalation paths multiply. It looks like better management. But underneath: routine execution requires alignment. App…
Cloud didn't just host applications. It shaped how they were built. Workloads moved first. Architecture adapted second. Data accumulated around platforms. Tooling locked into services. Teams optimize…
Cloud migration was not an architecture decision. It was capital allocation disguised as strategy. Startups moved because servers competed with hiring. Enterprises moved to convert CapEx to OpEx. Bot…
CBT, mindfulness, and deliberate practice were designed for the human mind — they should be the architectural blueprint for operational AI. When AI moves into real work — drafting contracts, approvin…
Growth hides structural problems. Revenue goes up. Product expands. Teams scale. It looks like success. But underneath, the system is fracturing. Product fragments into disconnected features. Deli…
After watching dozens of enterprise AI implementations, the pattern is clear: the technology works. The governance doesn't. Companies are nailing the technical stack. Models perform. Infrastructure s…
Most companies don't fail because the technology doesn't work. They fail because they don't change how they operate. I've seen this pattern across $600M businesses and failed startups. The technolog…
Why AI pilots never become operational systems. The problem is usually not the code or the model. The breakdown starts when trying to scale: - fragmented workflows - duplicated automation - isolated…
Most organizations are deploying AI the way they deployed software in 2005. New tool. Faster output. Same controls. Same accountability structure. Same governance assumptions. That is the category e…
Everyone talks about edge computing as a technology shift. It's not. It's a control shift. When decisions move closer to the data, three things become critical: Latency matters. Ownership matters. C…
Organizations don't operate according to strategy. They follow structure. Here's the loop that actually drives execution: Capital concentrates control. The money decides who makes decisions. Contro…
Most operating models were built for predictable change cycles. Plan, budget, execute over 12-18 months. Get approval through committees. Scale gradually. AI doesn't work that way. I watched a Fortu…
Most organizations survive because their systems are slow. Manual processes hide misalignment. People compensate for broken structures. Time absorbs mistakes. AI removes that buffer. Decisions happ…
Most companies think they have a data problem. They don't. They have a control problem. Data reflects how decisions are made. If ownership is unclear, data fragments. If incentives conflict, data get…
Most AI programs don't fail because the models are bad. They fail because no one knows who owns the decision. The model produces an output. The business overrides it. Accountability stays unclear. So…
Transformation rarely fails because the technology doesn’t work. It fails because existing structures lose relevance. And organizations defend the structures they know. SELLING TRANSFORMATION IN A …
Partnerships promise scale. Distribution. Market access. PARTNERSHIP IS NOT A SALES STRATEGY We built a commercial model around large enterprise partners. The logic was straightforward: global rea…
Most transformation plans assume execution speed will remain constant. But capital changes how organizations make decisions. And decision tempo determines whether transformation actually happens. C…
Automation rarely fails because of technology. It fails because of authority. The disruption is not the algorithm. It is who no longer makes the decision. AUTOMATION RARELY FAILS BECAUSE OF TECHNOL…
Execution often fails even when the strategy is sound. The market opportunity is real. The technology works. STRUCTURAL REALITY I have worked inside large enterprises, venture-backed companies, pr…
Notes written for a Perspective, one of the developed positions on enterprise execution.
No notes published in this group yet.
Evidence-led notes on the dynamics of the Australian AI market.
In April, EY announced it's embedding a multi-agent framework into EY Canvas across 160,000 audit engagements and 150+ countries. EY wrote: "Intelligent agents execute defined tasks, while auditors r…
The series on Staff and digital operators, in reading order, then related notes.
1 of 9 parts published, in reading order
Part 2 of 9
2/9 — Operational Visibility: "Autonomy without visibility is delegation without control." If an AI system is doing real work, the operator needs to see what changed, what moved, what is waiting, and…
Personal notes and observations.
AI is changing the shape of the workforce. Fewer people will sit at the centre of execution, while demand for their skills grows: judgment, architecture, context, control and verification. That is t…
Upcoming
Operator notes
(Oct 20, 2026)
Futurum Group reports that 46.9% of enterprise AI budgets are now running over plan. That is a different failure from the usual ROI story. The progra...
Perspectives
(Oct 21, 2026)
59% of APAC enterprises expect to increase AI investment by more than 25% this year, according to Digital Realty. Logicalis reports that only 16% of ...
The Staff series · Part 3 of 9
(Oct 23, 2026)
3/9 — Institutional Memory: "Most AI starts every conversation surprisingly close to zero." Not good enough for something meant to operate alongside ...
Operator Notes are one layer of BdG Advisory's Intelligence work — alongside Signals, Trends and Perspectives. Signals are observations. Trends link them into derived intelligence. Operator Notes and Perspectives are content artifacts with provenance — a note may be developing thinking, a perspective is a stated position, and either can draw on the others without having to.