Sun 5 April 2026

AI Is Collapsing the Software Stack: Platforms Are Becoming the Entire Value Chain - The Xero–Anthropic integration is a structural shift in competitive advantage. By embedding a frontier reasoning model like Claude directly into a system of record with deep, proprietary financial data, Xero is collapsing entire categories of adjacent software and services into its core platform. We have seen this playbook before: when distribution, data, and intelligence converge, standalone tools become features, and features become invisible. The implication is stark—competitive boundaries are no longer defined by functionality, but by control of data ecosystems and the ability to operationalise AI natively within them. For incumbents and challengers alike, this reframes strategy. You cannot compete horizontally against platforms that combine scale, data, and embedded intelligence. The defensible position shifts to vertical depth—owning highly specialised workflows, edge cases, and domain-specific contexts that generalised AI layers will not prioritise. Crucially, AI is no longer a differentiating feature; it is table stakes infrastructure. The winners in this next phase will be those who move fastest to deploy specialised, deeply integrated AI agents within their niche—effectively building micro-platforms of expertise that sit beyond the reach of horizontal consolidation.

AI vibe coding tools are subversive in that it empowers not IT staff to build and ship code. But that can lead to a lot of challenges for security, privacy and maintainability. Filev's account at Zencoder illustrates a pivotal shift in software development, where AI agents have drastically reduced implementation costs, empowering product managers and designers to bypass traditional handoffs and directly ship features—evident in a PM building a waiting-game widget in a day or a designer tweaking IDE plugins in real time. This disrupts the org chart by eliminating coordination bottlenecks once justified to safeguard engineering bandwidth, fostering faster decision velocity and compounding effects like sharper specifications and heightened ownership. While the narrative aligns with broader AI-driven democratization trends, it warrants skepticism: Filev's optimism for scaling in complex brownfield environments overlooks potential pitfalls, such as code quality inconsistencies, security vulnerabilities from non-engineer contributions, or the risk of siloed innovations that fragment enterprise coherence, especially in regulated industries where accountability remains paramount. Key points: • AI agents collapsed implementation costs, shifting bottlenecks from engineering to decision-making. • PMs and designers now ship directly, reducing tickets, handoffs, and translation layers. • This enables pursuit of low-priority but high-impact ideas, like personality-adding features. • Compounding effects include sharper specs, fewer iterations, and a culture of universal building. • Implications extend to larger organizations, closing the gap between intent-holders and builders. Expert question (counterfactual): What if the surge in non-engineer code contributions leads to unmanageable technical debt or compliance risks in highly regulated sectors, undermining the promised velocity gains?

This NBER working paper, drawing on a survey of nearly 750 executives, reveals a mixed landscape for AI's impact on productivity and labor markets, underscoring significant firm-level heterogeneity in adoption rates—over half have invested, yet smaller entities lag behind. Productivity gains are evident and sector-specific, particularly in high-skill services and finance, driven by total factor productivity enhancements via innovation and demand rather than mere capital intensification, with projections for acceleration in 2026. However, a notable productivity paradox emerges, where executives' perceptions of gains outpace measurable outcomes, possibly attributable to lagged revenue effects—a claim that warrants scrutiny amid potential optimism bias in self-reported data. On employment, aggregate near-term declines appear minimal, though reallocation is underway, favoring skilled technical roles over routine clerical ones, with larger firms eyeing reductions while smaller ones anticipate growth; this suggests transitional disruptions rather than outright catastrophe, but long-term dynamics remain uncertain. Key points: • Over 50% of firms have invested in AI, with heterogeneity favoring larger entities. • Productivity gains are positive, sector-varying, and expected to rise in 2026, led by high-skill services and finance. • Perceived productivity benefits exceed measured ones, linked to delayed revenue realization. • Minimal aggregate job losses anticipated short-term, but shifts from clerical to technical roles are evident. Expert question (counterfactual): What if the productivity paradox stems not from revenue lags but from executives overestimating AI's standalone contributions, conflating them with concurrent digital transformations or market recoveries?

A critical imbalance in corporate AI strategies, with 93% of budgets allocated to technology and a mere 7% to workforce preparation, as per Deloitte and Wharton analyses, potentially creating bottlenecks and resistance that undermine adoption. Experts like Eric Bradlow and Lara Abrash warn that this tech-centric approach exacerbates organizational weaknesses, particularly among middle managers, while neglecting human skills like curiosity, emotional intelligence, and divergent thinking essential for AI augmentation. Leadership must evolve from 'pathfinding' to 'wayfinding' amid uncertainty, fostering 'bridger' roles to integrate tech and people. However, the narrative's doomsday tone may overstate risks; historical tech shifts suggest adaptive firms could thrive, though the revenue upside from co-intelligence—estimated at $6 billion for a $60 billion company—remains speculative without broader evidence of successful human-AI integration. Key points: • Companies allocate 93% of AI budgets to tech and only 7% to workforce development, leading to adoption failures. • Middle managers often resist AI due to lack of preparation, creating bottlenecks in workflows. • Human strengths like curiosity, emotional intelligence, and divergent thinking are irreplaceable for effective AI use. • New leadership requires 'wayfinding' and 'bridger' roles to navigate AI uncertainties. • AI could drive $6 billion in annual revenue growth for a $60 billion firm through enhanced productivity and innovation. Expert question (counterfactual): What if the projected revenue gains from AI-human collaboration fail to materialize due to persistent workforce resistance, forcing companies to revert to cost-cutting and further erode employee trust?

WSJ reporting AI is creating new jobs - AI is proving to be a job creator as well as a disruptor. LinkedIn data show about 640,000 new U.S. jobs tied to AI between 2023 and 2025, including roles such as Head of AI, AI Engineer, AI Consultant, AI/ML Researcher, Forward-Deployed Engineer, and Data Annotator, not counting the surge in data center construction work. These jobs span technical, strategic, and operational work across industries including finance, healthcare, and manufacturing, where people are helping both train AI systems and help others use them effectively.

https://arxiv.org/abs/2604.01363

Anthropic's back-to-back leaks—first a draft blog post unveiling the potent 'Mythos' or 'Capybara' model with heightened cybersecurity risks, now 500,000 lines of Claude Code source code—expose troubling lapses in operational security at a firm pioneering AI safety. While Anthropic downplays the code exposure as mere 'human error' in packaging, without sensitive data loss, experts rightly highlight its value: the agentic harness code reveals proprietary guardrails, tool integrations, and internal APIs, potentially enabling competitors to reverse-engineer enhancements or adversaries to probe safeguards. This pattern, echoing a February 2025 incident, undermines Anthropic's safety-first ethos, especially as Capybara promises unprecedented capabilities like zero-day vulnerability detection that could be dual-use. Skeptically, their assurances of robust processes ring hollow amid repeated misconfigurations, signaling a need for stricter release protocols in an industry racing toward AGI amid escalating risks. Key points: • Anthropic leaked 500,000 lines of Claude Code source code via NPM due to human error, exposing agentic harness details. • The leak follows a prior exposure of 'Mythos/Capybara' model info, described as more advanced than Opus with major cybersecurity implications. • Experts warn the code could aid competitors in replicating features or attackers in bypassing safeguards. • No customer data was compromised, but internal architecture insights were revealed, potentially informing open-source alternatives. Expert question (counterfactual): What if these leaks were not mere accidents but subtle strategic disclosures to accelerate industry-wide safety discussions on dual-use AI risks?

Thomas Mulligan, a former CIA case officer and RAND researcher, argues persuasively that the proliferation of generative AI, particularly large language models, will paradoxically elevate the role of human intelligence (HUMINT) in espionage. As AI democratizes disinformation—enabling agents to fabricate convincing backstories or false intelligence—it erodes trust in digital channels, pushing operatives toward analog methods like dead drops and face-to-face encounters. While AI offers tools for pattern recognition in surveillance or psychological manipulation, Mulligan warns it also heightens risks for human handlers, making counterintelligence more vital. This perspective revives Cold War-era spycraft, but skeptics might question whether it underplays AI's potential to automate even interpersonal deceptions, potentially rendering human elements as vulnerable relics in an increasingly synthetic intelligence landscape. Overall, Mulligan's analysis underscores a timely shift: in an era of deepfakes, the human touch may indeed be the ultimate verifier of truth. Key points: • AI enables easier fabrication of false intelligence, increasing the need for human verification in HUMINT. • Digital communications lose trustworthiness due to AI-generated disinformation, favoring old-school spycraft like dead drops. • AI tools can aid spies in manipulation and data analysis but also pose detection risks for operatives. • Human intelligence officers will remain essential as AI complicates distinguishing fact from fiction. Expert question (counterfactual): What if rapid advancements in AI-driven behavioral prediction and real-time deepfake countermeasures ultimately automate HUMINT tasks, making human spies as obsolete as they were feared to be?

AI-mediated search is fragmenting public information ecosystems, compelling governments to overhaul digital strategies to preserve policy efficacy and equity in an era of uncontrolled summarization. The UK Department for Education reports surging AI-driven traffic alongside declining direct visits, highlighting risks of narrowed user understanding and disadvantaged access for less confident individuals. "We now need to design with the expectation that much of what we publish will be read indirectly, atomised, summarised or reinterpreted by systems we don't control," writes Mark Edwards, head of design for the UK government's Department for Education.

We always need to listen to Ethan. He is bang on in that firms must treat AI as a fundamentally novel technology rather than routine IT automation to mitigate risks and harness opportunities, avoiding adverse impacts on operations and workforce stability. 'It is a strange technology with both risks & opportunities that need to be discovered. Pretending AI works like normal IT automation can result in bad outcomes for companies & their employees.'

As AI agents increasingly automate code writing and review, businesses face heightened risks from overreliance without robust governance, potentially amplifying security vulnerabilities and operational blind spots in software development. Anthropic's new Code Review feature employs specialized AI agents to detect bugs and prioritize issues, but experts stress the need for adapted workflows, human oversight, and observability to maintain accountability and prevent compounded errors.

A clear admission that a multi-horse betting strategy is required for underlying foundational models.

Alibaba's Qwen3.6-Plus advances China's AI competitiveness by targeting practical real-world agents, intensifying global rivalry and accelerating automation in business operations. This model builds on prior iterations to enable more autonomous AI applications, potentially reshaping labor markets and investment priorities in AI infrastructure. Alibaba's release of Qwen3.6-Plus signals accelerating Chinese innovation in AI agents, heightening geopolitical competition and pressuring Western firms to advance practical AI deployment for real-world applications. The model focuses on building autonomous agents capable of handling complex, real-world tasks, potentially reshaping labor markets and enterprise efficiency.

Interesting Microsoft is behind in building leading frontier models as it lacks the necessary compute! "Microsoft has unveiled its latest midsized AI model as it seeks to gain a foothold in the technology, but chief Mustafa Suleyman said the tech giant still lacks the computing power needed to build cutting-edge systems. The software giant on Thursday released a speech transcription model that Suleyman called the most advanced of its kind, as it steps up efforts to close in on rivals and reduce its reliance on OpenAI. Microsoft has yet to release large language models capable of competing in more sophisticated areas such as coding and text generation, where it lags market leaders Anthropic, Google and OpenAI."

Benjamin Laker's column astutely delineates the boundaries of AI in managerial practice, advocating its use for accelerating rote tasks like data synthesis and drafting while cautioning against outsourcing human-centric judgments on values, relationships, and trust. This balanced perspective counters the hype surrounding AI as a panacea for productivity, emphasizing that overreliance risks eroding critical thinking and emotional intelligence—core to effective leadership. However, Laker's framework assumes managers possess the self-awareness to discern these boundaries consistently, a skepticism warranted given evidence of cognitive biases in tech adoption. By positioning AI as a tool rather than a proxy, the piece promotes deliberate integration, potentially fostering more resilient organizational cultures amid rapid technological shifts, though it underplays the need for systemic training to prevent unintended delegation of authority. Key points: • Use AI for mechanical tasks like summarizing reports or outlining agendas to free up time for strategic sensemaking. • Avoid AI in decisions involving trust, values, or relationships, such as delivering feedback or assessing candidate resilience. • Treat AI outputs as raw material requiring human editing to maintain judgment and accountability. • Instruct AI to generate counterarguments to challenge biases and broaden decision-making perspectives. Expert question (counterfactual): What if advancing AI capabilities in emotional intelligence and contextual nuance render the distinction between mechanical and human tasks obsolete, forcing managers to redefine leadership entirely?

We find this all the time - the limiter to AI adoption is people. Organizational readiness, not AI model quality, determines the speed and success of enterprise AI deployments, positioning firms that invest in adaptive processes and leadership for competitive advantage in the AI economy. Stanford's Digital Economy Lab analysis of 51 cases shows identical technologies delivering outcomes from weeks to years, underscoring the need for internal change to capture AI value.

ServiceNow's push into agentic AI frames autonomous workflows as the next frontier beyond copilots, promising to redefine enterprise work through self-acting software, but only with rigorous governance to manage risks like data access and outcome drift. Executives advocate treating AI agents as a distinct identity class—neither fully human nor machine—equipped with least-privilege permissions, starting implementations in controlled IT domains like email triage to demonstrate tangible ROI, such as slashing resolution times by 13%. Tools like the AI Control Tower enable runtime monitoring and traceability, ostensibly safeguarding PII and ensuring alignment with intent. While this narrative aligns with ServiceNow's platform-centric strengths and addresses real security fault lines, it warrants skepticism: the hype around productivity gains may overlook integration complexities and the historical struggle of enterprises to monetize AI, potentially positioning governance as a compliance burden rather than a velocity booster. Key points: • Treat AI agents as a new identity class with distinct, scoped permissions for data access. • Begin agentic AI deployments in narrow IT use cases to measure ROI, like reducing task times from days to minutes. • Employ AI Control Tower for logging actions, tracing outcomes, and monitoring drift to maintain security and traceability. • Prioritize governance and security to enable scaling, as lapses could stall AI programs amid CIO pressures for value realization. Expert question (counterfactual): What if robust governance, while mitigating risks, inadvertently creates bureaucratic hurdles that hinder the rapid experimentation needed for agentic AI to truly outpace human workflows in diverse enterprise domains?

BCG's exploration of agentic-commerce futures presents a forward-looking framework for marketers navigating AI-driven shopping agents, outlining four scenarios: an open bazaar of seamless transactions, brand resurgence via proprietary data ecosystems, super apps consolidating services, and creator-led authenticity emphasizing human touch. While the analysis astutely reduces these to core imperatives—machine discoverability and brand desirability—it risks oversimplifying the interplay of regulatory hurdles, privacy concerns, and technological fragmentation that could derail such visions. The pivot from SEO to 'answer-engine optimization' is pragmatic, urging brands to prioritize structured data and conversational relevance, yet it assumes agents will uniformly favor discoverable, desirable content without accounting for biases in AI training or monopolistic platform dominance, potentially favoring incumbents over innovators. Key points: • Four agentic-commerce scenarios: open bazaar, data ecosystems for brands, super apps, and creator authenticity. • Core requirements: enhance machine discoverability and brand desirability. • Shift strategy from SEO to answer-engine optimization for AI interactions. Expert question (counterfactual): What if AI agents prioritize user privacy and ethical sourcing over discoverability, forcing brands to invest more in transparent, verifiable desirability rather than optimized data feeds?