# Shane Paola — Full Site Text > One file containing every public page of shanepaola.com in plain markdown, for LLM ingestion. > Canonical site: https://www.shanepaola.com =========================================================== # / (Home) =========================================================== # Shane Paola — GTM Advisory & Sales Leadership Executive sales leadership and go-to-market advisory for B2B tech companies. 24 years of experience, $850M+ in revenue built. Based in the UK, working globally. ## About Shane Paola is a senior GTM executive and independent advisor specialising in scaling enterprise tech go-to-market functions. Past leadership roles span Cisco, EMC, Meraki, ThousandEyes, Plume, and SamKnows. The practice focuses on three areas: Board Advisory, Fractional Leadership, and Thought Leadership. ## The challenge most GTM teams face Most enterprise tech GTM teams oscillate between three failure modes: they freeze (waiting for the perfect strategy), they flail (running a dozen plays at once with no system), or they react (jumping from quarter to quarter on whatever the board last asked for). The "Zoom Out, Zero In" methodology exists to break that pattern. ## Zoom Out, Zero In™ — the methodology A four-stage flow used in every engagement: 1. **Clarity** — get the unvarnished picture of where the GTM function is today. 2. **Focus** — identify the small number of moves that actually shift the trajectory. 3. **Align** — make sure leadership, marketing, sales and CS are pulling the same way. 4. **Execute** — sequence the work so it compounds quarter over quarter. ## The 6 GTM Health Dimensions Three upstream dimensions (Market Positioning, ICP & Targeting, Value Proposition) and three downstream dimensions (Sales Motion, Revenue Operations, Customer Lifecycle). The free diagnostic at gtm.shanepaola.com scores an organisation against all six. ## Services - **Board Advisory** — strategic GTM input for founders, CEOs, and boards navigating scale, pivots, or exits. - **Fractional Leadership** — interim CRO / VP Sales engagements for companies between full-time hires or rebuilding the function. - **Thought Leadership** — keynote speaking, workshops, and writing (including the AI-First GTM Operating Model paper and the weekly AI News recap). ## Selected experience - **ThousandEyes** — first international GTM hire, scaled to acquisition by Cisco. - **Meraki** — early EMEA leadership, scaled to acquisition by Cisco. - **Plume** — built EMEA enterprise from zero. - **SamKnows** — Head of Sales through a multi-year growth phase. - **Cisco / EMC** — enterprise sales leadership across multi-hundred-million-dollar territories. Combined: $850M+ in revenue built across 24 years. ## Contact - LinkedIn: https://linkedin.com/in/shanepaola - Book an intro: https://cal.com/shane-paola-msv9dn - Free GTM diagnostic: https://gtm.shanepaola.com - Email: shane@shanepaola.com =========================================================== # /ai-first-gtm-operating-model =========================================================== The AI-First GTM Operating Model — Shane Paola A Shane Paola Working Paper · v2.0 · May 2026 An Operating Framework # The AI-First GTM Operating Model A framework for transforming enterprise tech go-to-market functions in the age of AI. Written for organisations rewiring the plane while it is flying. Shane Paola Senior GTM Executive & Independent Advisor Version 2.0 Living Document Contents # What's inside A framework, five pillars, and a practical operating model. Read it in order. The sequence of pillars is itself part of the argument. 01The Premise04 02The Five Pillars05 03Pillar One. People07 04Pillar Two. Environment09 05Pillar Three. Process10 06Pillar Four. Data Foundation13 07Pillar Five. Tools and Agent Infrastructure14 08Function-by-Function Breakdown16 09Org Structure Implications19 10What This Is Not21 11What Comes Next22 Shane Paola Advisory02 01 · The Premise Section 01 # The Premise We are getting too wrapped up in the headlines. Mass layoffs, AI disruption, re-orgs. The noise is deafening. Strip it back, though, and what is actually happening is a structural reset. Some of those cuts are genuine AI-driven efficiency plays. Others are cost offsets for the enormous capital required to build out the data centre and GPU infrastructure powering the next wave of AI platforms. Either way, the net result is the same. The way enterprise tech businesses run their go-to-market functions is changing, and the gap between the organisations that get this right early and the ones that do not will widen fast. The real opportunity is not just cost reduction. It is supercharing the resources you currently have. We saw it with SaaS and the cloud. When mass disruption hit one side of the industry, it almost always overcompensated on the other through new businesses, new categories, and a new generation of entrepreneurship. AI is setting up the same dynamic. The organisations who read this correctly will build a structural advantage. The ones who protect legacy ways of working and refuse to unlearn will fall behind. The divergence has already begun. This document is built for a specific audience: enterprise tech companies with existing GTM functions (sales, channel, SDR, revenue operations) who are trying to transform, not start from scratch. A greenfield AI-native company is a different problem. Here, the challenge is bigger. You are rewiring the plane while it is flying. You have legacy systems, embedded habits, existing headcount, and a number to hit. That context shapes everything that follows. The Core Argument AI-first GTM transformation is not a cost reduction exercise. It is a competitive differentiation play. The enterprise tech businesses that build genuinely AI-native go-to-market functions in the next 18 to 24 months will have a structural advantage in pipeline coverage, sales cycle velocity, and customer intimacy that laggards will struggle to close. Shane Paola Advisory04 02 · The Five Pillars Section 02 # The Five Pillars An AI-first go-to-market operating model breaks down into five interdependent pillars. Get one wrong and it undermines the others. Build them in the right sequence and what emerges is a genuinely different kind of GTM engine. Faster, more intelligent, and structurally more competitive than what most enterprise tech businesses are running today. People. Mindset before everything else. Environment. The culture that makes it work. Process. Mapping where the time actually goes. Data Foundation. The bedrock everything else runs on. Tools and Agent Infrastructure. How you bring it to life. The order matters. Most organisations try to start with tools. That is the wrong entry point. You cannot buy your way into an AI-first GTM model. You have to build your way there. Figure 01 ### The Five Pillars: sequence of build, not menu of choice Build left to right. The dashed loop reflects the reality that tools and agents change what "the right people" looks like, and the cycle re-runs. Shane Paola Advisory05 03 · Pillar One. People Section 03 · Pillar One # People 01People 02Environment 03Process 04Data 05Tools ## 3.1   Mindset is the differentiator Getting the right people on the bus is the single most important step. It is also the one most organisations underinvest in before they move to the more tangible work of process redesign and tool selection. To be direct, this is not a demographic conversation. It has nothing to do with age, gender, or years of experience. A rep with 20 years in enterprise tech can be exactly the right profile for this new world if they have the right mindset. A 26-year-old can be entirely wrong for it if they are rigidly attached to a specific playbook and unwilling to challenge it. The profile you are looking for is defined by a specific set of orientations: An optimistic view on AI. Not blind enthusiasm. Genuine curiosity, and openness to what is possible. A willingness to unlearn. Question ways of operating that have delivered results historically, and let go of the ones that no longer serve. A genuine growth mindset. Actually invest time to build new capabilities, rather than paying lip service to the concept. Comfort with ambiguity. The playbooks of the last decade are not dead yet, but they are changing fast. The ability to operate without a complete map is essential. The militant playbook junkies will find this uncomfortable. Not because the fundamentals of great selling are gone (they are not), but because the surrounding infrastructure of how great selling gets done is being disrupted entirely. The core skill remains constant: building trust, understanding customer pain, orchestrating complex deals. Everything else is changing. Shane Paola Advisory07 03 · Pillar One. People ## 3.2   Talent segmentation One of the most practical early steps before redesigning anything else is running a talent segmentation exercise against your existing team. Categorise people into three buckets: AI Optimists. Already experimenting. Naturally curious. Probably using AI tools outside work already. These are your change agents. Deploy them first. Let them build real-world credibility for what is possible inside your organisation. AI Pragmatists. Open to it but need proof before they commit. Not resistant, just not self-starters on adoption. This is almost certainly your largest segment. They will follow the optimists if the environment is right. AI Resistors. Actively sceptical or ideologically opposed. Some will shift with the right support and time. Some will not. Be honest with yourself about which is which, early. This segmentation determines your change management sequencing, not your headcount decisions. The goal is to move the pragmatists into the optimist camp, rather than to clear out the resistors. Figure 02 ### Talent segmentation: population shape and direction of change The work is to shift the centre mass, rather than to purge the left tail. Population shares are typical patterns, not benchmarks. Shane Paola Advisory08 04 · Pillar Two. Environment Section 04 · Pillar Two # Environment 01People02Environment03Process04Data05Tools ## 4.1   Culture as the operating system Assuming you have the right people on the bus, you need to create the right environment for them to thrive. This is fundamentally a learning and experimentation environment. People should be able to implement ideas quickly, test them in the field, and iterate without fear. The culture has to do more than tolerate that approach. It has to actively celebrate it. A lot of enterprise tech businesses are poor at this. They have built cultures optimised for the predictability of the SaaS era: forecast accuracy, QBR rituals, stage-gate approval processes. None of those disappear overnight. They need to coexist with a very different operating rhythm for the teams building the new model. Leadership has to be fully on the same wavelength. You cannot ask teams to experiment and move fast if the leadership above them is punishing failure and demanding certainty. The culture signals have to be consistent from the top. Leaders who create psychological safety for people to bring ideas to the table, try things, fail quickly, and adapt: that is what the environment needs to look like. ## 4.2   Four structural conditions for experimentation Four structural conditions tend to determine whether an AI-first culture actually takes root or just gets paid lip service: Dedicated experimentation time. Teams need protected time to learn and build. If AI adoption gets squeezed between quota calls and pipeline reviews, it does not happen. The best-performing GTM teams treat AI fluency like a core sales skill: it gets trained, practised, and measured. Shared learning infrastructure. A living, searchable, accessible record of GTM AI experiments. Prompts that work. Agents that saved time. Processes that got automated. Not a monthly all-hands update. An actual institutional memory that compounds over time. Leader modelling. Leaders who demonstrate the behaviours they are asking for. Not deck deep-dives on AI strategy. Frontline leaders showing their team, specifically, how they personally use AI in their daily workflow. That is what actually moves culture. Failure tolerance metrics. If the only metrics that matter are quota attainment and activity volume, an AI-first culture will never take root. Lightweight signals that reward experimentation, adoption, and improvement need to sit alongside the output metrics. Shane Paola Advisory09 05 · Pillar Three. Process Section 05 · Pillar Three # Process 01People02Environment03Process04Data05Tools ## 5.1   The high-yield vs. low-yield map This is where the transformation gets operationally real. Once the right people are in place and the right environment exists, the next step is a deep and honest process audit. Not designing a new process on a whiteboard, but mapping what the team actually does today. Walk through a typical week with your team. Document honestly where the time is going. Then categorise every activity against a simple framework: High-yield activities. The effort-to-output ratio is strongly positive. Time spent here compounds. Customer-facing time, relationship deepening, deal navigation, discovery, executive engagement. Anything that moves a customer relationship or a deal materially forward. Low-yield activities. High effort, low or inconsistent output. Activities that happen because they have to, not because they create disproportionate value. Research, admin, reporting, presentation building, data entry, internal coordination overhead. The goal here is not to eliminate the people doing low-yield activities. The goal is to eliminate the friction so those people can stop doing them. That framing matters. This is not an optimisation-out-of-headcount exercise. It is a supercharge play. Removing the work that drains energy and produces subpar results precisely because people resist doing it naturally, and delegating that work to AI infrastructure that does it without friction and often with better accuracy. Shane Paola Advisory10 05 · Pillar Three. Process ## Where time goes today vs. where time should go The transformation is a reallocation, not a reduction. AI infrastructure absorbs the low-yield work. The rep's week refills with high-yield activity. Figure 03 ### High-yield vs. low-yield: what AI infrastructure absorbs Bar lengths represent illustrative effort and time today. The framing is reallocation: same headcount, structurally more effective. Shane Paola Advisory11 05 · Pillar Three. Process ## 5.2   Priority automation targets From experience running GTM teams at scale, the highest-value automation targets tend to cluster around: Pre-meeting customer research and executive briefing document generation QBR deck preparation and data aggregation Presentation building (on-brand, customised, and consistent) CRM data entry and pipeline hygiene Reporting and dashboard generation Proposal first drafts and legal or commercial playbook navigation Competitive intelligence monitoring and synthesis Account onboarding coordination and post-close internal handoff Win/loss pattern analysis across the closed pipeline Partner pipeline reconciliation for channel teams ## 5.3   What high-yield looks like when the shift works When the process transformation lands correctly, what the rep gets to do with their time changes fundamentally. They show up to customer meetings informed and prepared, rather than having spent 45 minutes the night before building a brief of questionable accuracy. They are not trapped in their CRM at end of week trying to reconstruct what happened in five calls. They are not building their QBR deck on a Sunday. That time is reallocated to the work that actually matters: sitting with customers, understanding their businesses in real depth, learning the financial mechanics that make a proposal land or die, building multi-threaded relationships across the buying committee. The Reframe That Matters The rep stops being a vendor pushing a product and becomes a collaborative partner locked arms with the customer on a real problem. Your technology is the enabler. Revenue and bookings are the lagging indicator of a relationship done right, rather than the primary driver of the interaction. Shane Paola Advisory12 06 · Pillar Four. Data Foundation Section 06 · Pillar Four # Data Foundation 01People02Environment03Process04Data05Tools ## 6.1   The non-negotiable bedrock Here is where something often gets skipped because it is unglamorous. None of the above works if the data foundation is broken. The right people, the right culture, the most sophisticated agent stack in the market: if the underlying data is dirty, inconsistent, or siloed, the intelligence layer built on top of it is worthless. The data foundation work has to happen in parallel with the process mapping and people work. It does not need to be perfect before the rest begins, but it needs to be directionally solid and consistently structured across all GTM functions. Without it, the result is automated garbage. The organisation loses trust in the AI layer fast. ## 6.2   What a clean data foundation requires In practical terms for a GTM AI-first transformation, a clean data foundation comes down to four requirements: Unified customer and account data. A single source of truth for account hierarchy, contacts, relationship history, and engagement data. The discrepancies between CRM, marketing automation, and CS platform that tell different stories about the same customer have to be resolved. Structured activity capture. Meeting notes, call recordings, email summaries, and interaction data feeding back into the system consistently. Ideally AI-assisted capture, so reps are not manually logging and the data is complete by default. Clean pipeline mechanics. Stage definitions that actually mean something, deal hygiene standards enforced by process or by an agent, and closed-loop data on wins and losses that can be learned from at scale. Consistent taxonomy. Segment definitions, product categories, and use case classifications agreed across sales, marketing, and CS, applied consistently in every system. This sounds elementary. It almost never is. The Investment That Pays Compound Interest The highest-leverage investment most enterprise GTM teams can make right now is not another AI tool. It is three to six months of focused data hygiene and structure work that makes every subsequent AI investment dramatically more effective. The temptation to skip this and go straight to deployment is strong. Resist it. Shane Paola Advisory13 07 · Pillar Five. Tools & Agent Infrastructure Section 07 · Pillar Five # Tools and Agent Infrastructure 01People02Environment03Process04Data05Tools ## 7.1   The operating stack philosophy Once the people are right, the environment is set, the processes are mapped, and the data foundation is intact, then it is time to talk about tools. Not before. The instinct in most organisations is to reverse this order. To go tool shopping before the foundations are in place. That is how expensive AI stacks end up unused, because the data feeding them is unreliable, or the processes were never redesigned to actually leverage the new capability. The philosophy here is ruthless simplification. Do not try to cover everything with dozens of disconnected tools. Build everything connected underneath and leave the rep with the simplest possible interface on top. The direction emerging from the most advanced deployments is building the tech stack around an agent framework where CRM, communication platforms, data sources, and productivity tools are all connected via model context protocols (MCPs), with agents sitting in the middle orchestrating across all of them. The rep does not need to be a power user of ten different platforms. They interact through a single conversational interface (Slack, Teams, WebEx, whatever the business communication platform is) and a swarm of agents do the work behind the scenes. ## 7.2   The three-layer agent architecture The agent infrastructure for a GTM team breaks naturally into three functional layers: ### Layer 1. Intelligence Agents (always on) These run continuously in the background, monitoring and synthesising so no rep ever starts their day from zero: Account intelligence agent. Monitors news, earnings calls, leadership changes, and product announcements for named accounts. Competitive intelligence agent. Tracks competitor moves, pricing, positioning, and market reviews across the competitive set. Pipeline health agent. Monitors deal progress, flags at-risk opportunities, and surfaces coaching moments. Shane Paola Advisory14 07 · Pillar Five. Tools & Agent Infrastructure ### Layer 2. Workflow Agents (triggered by events or schedules) These activate when something happens or on a defined timetable: Pre-meeting preparation agent. Generates customer briefing, agenda, context, and suggested questions 24 hours before any logged meeting. Post-meeting processing agent. Captures notes, extracts action items, updates CRM, and drafts follow-up communications. QBR preparation agent. Aggregates performance data, surfaces narratives and risks, and generates initial deck structure. ### Layer 3. Conversational Agents (on-demand) These respond to requests in real time through the chat interface: Research agent. Answers questions about accounts, contacts, markets, and competitors on demand. Content agent. Drafts emails, proposals, executive summaries, and presentation content. Coaching agent. Reviews deal plans, challenges assumptions, and provides deal strategy support. The key design principle: Layer 1 agents feed Layer 2 agents, and both feed the conversational Layer 3 agents with context. The rep interacts primarily with Layer 3. The intelligence and workflow work happens transparently underneath. Figure 04 ### The three-layer agent architecture Layer 1 feeds Layer 2. Both feed Layer 3 with context. The rep only ever interacts with Layer 3. Everything beneath is invisible. Shane Paola Advisory15 07 · Pillar Five. Tools & Agent Infrastructure ## 7.3   Stack design principles Regardless of which specific tools are deployed, five non-negotiable design principles apply: Single source of truth. Every tool in the stack writes back to the same underlying data layer. Fragmented data stores kill agent effectiveness at scale. Conversational first. The primary interface for the rep must be conversational. Agents that require dedicated UIs and complex navigation will not be adopted. Invisible complexity. The sophistication of the system should be invisible to the end user. The experience should feel simple, even when what is behind it is not. Measurable output. Every agent and workflow needs a measurable output. If you cannot measure whether it is saving time or improving quality, you cannot improve it. Modular architecture. Build so components can be swapped without rebuilding everything. The agent and tool landscape is moving too fast to lock into a single vendor for the whole stack. Shane Paola Advisory16 08 · Function-by-Function Breakdown Section 08 # Function-by-Function Breakdown ## 8.1   Enterprise sales The direct sales function is where the impact is most immediately visible. The core transformation is time reallocation, from admin-heavy, low-yield activity toward high-touch, high-value customer engagement. With the right agent infrastructure in place, a senior enterprise rep should be spending 70% or more of their working week in direct customer engagement, deal navigation, and relationship development. Most enterprise sales teams are a long way from this today. Benchmark · Recoverable Capacity Well-configured AI infrastructure can recover 8 to 12 hours per rep per week from low-yield activities. Across a 20-person enterprise sales team, that is between 160 and 240 hours of reclaimed capacity per week. The question every GTM leader should be asking is what that does to pipeline coverage ratio. ## 8.2   SDRs The SDR function gets significant airtime in AI disruption conversations. The view here is that the role still exists, but it looks meaningfully different. The SDR who spends their day manually building lists, writing generic outreach sequences, and logging activity data faces real displacement pressure. Not through active headcount decisions, but through competitive market pressure. Organisations using AI are doing the same volume and quality of outreach at a fraction of the resource cost. The SDR who survives and thrives in this environment is the one who treats AI as leverage. They orchestrate multi-signal outreach, personalise at scale, and spend the bulk of their time on the high-touch conversations that AI cannot replicate. The calls that require genuine human judgment, empathy, and improvisation in real time. There is also a structural argument that the early-stage qualification work SDRs do today becomes an automated pre-pipeline function over time. If that happens, the human SDR resource gets redeployed toward more complex account development, partner-assisted pipeline, and enterprise qualification where human judgment is genuinely necessary. Shane Paola Advisory17 08 · Function-by-Function Breakdown ## 8.3   Channel and partner teams Channel reps have always carried a disproportionate admin load. Partner management, enablement, pipeline reporting, and co-selling coordination create a real time drain that sits on top of an already complex external-facing role. The AI-first model changes this. With agents handling partner pipeline reporting, competitive enablement content, co-sell briefing preparation, and partner-sourced opportunity tracking, the channel rep can redirect their time toward the work that actually drives sourced pipeline: executive relationships, strategic co-selling motions, and partner programme development. There is also an under-explored opportunity in AI-assisted partner segmentation and tiering. Most channel programmes have a long tail of partners who receive standard enablement but deliver disproportionately little pipeline. AI analysis of partner performance data, engagement patterns, and market coverage produces a more precise segmentation, allowing resource to flow toward partners with genuine growth potential rather than being spread uniformly across a base of hundreds. ## 8.4   Sales operations and revenue operations If the data foundation is clean, sales ops and rev ops transform from report-builders and CRM administrators into genuine strategic advisors to the business. The analytical capacity that previously required a team running queries and building dashboards gets automated. What remains is interpretation, recommendation, and programme design: work that requires human judgment and deep business context. The rev ops function in an AI-first GTM model is also naturally positioned to become the internal owner of the agent infrastructure. The team that owns configuration, optimisation, and performance measurement of the GTM agent stack. This is a natural extension of their existing role as system owners, and it gives them a genuinely strategic charter: building and operating the intelligence layer that the entire GTM function depends on. Shane Paola Advisory18 08 · Function-by-Function Breakdown ## GTM function transformation: before and after Each function reallocates time differently. The pattern is consistent. Low-yield activity collapses, high-yield activity expands, and the role gets meaningfully more strategic. Figure 05 ### Time split today vs. AI-first model, by function Splits are illustrative directional patterns observed across enterprise tech GTM teams, not benchmarked averages. The shape (heavy admin collapsing into customer time) is consistent. Shane Paola Advisory19 09 · Org Structure Implications Section 09 # Org Structure Implications ## 9.1   The case for a flatter structure Here is the contentious argument. When real-time, consistent data is accessible to everyone in the GTM organisation, and when agents are doing the monitoring, flagging, and synthesis work that managers have historically performed, the case for traditional middle management layers weakens. Not disappears. Weakens. A manager's traditional value-add has comprised three things: information aggregation and cascade, performance monitoring and coaching, and administrative coordination. AI handles the first and large parts of the third. Performance monitoring becomes increasingly self-serve. What remains is coaching and human development, which still matters enormously, but it does not necessarily require the same span of control ratios that traditional GTM organisations have been built on. ## 9.2   The new individual contributor profile What changes most fundamentally is the profile of the individual contributor. The rep in the AI-first era needs a new set of capabilities alongside the traditional skills: The ability to operate and direct a virtual team of agents, becoming an orchestrator of AI tools rather than merely a user of them Enough technical fluency to configure, prompt, and extract maximum value from AI infrastructure Strong business acumen and financial literacy, because they now show up to customers with better intelligence, and customer expectations of insight will rise accordingly An analytical mindset, with comfort working with data and acting on AI-generated recommendations alongside instinct and experience This profile shift has direct implications for hiring. The job description for an enterprise sales rep in 2026 looks materially different from 2021. AI fluency should be a baseline competency requirement, not a nice-to-have, in the same way CRM proficiency became a baseline expectation a decade ago. GTM leaders not including AI fluency in hiring criteria are building the wrong team for the next cycle. Shane Paola Advisory20 09 · Org Structure Implications ## Traditional GTM org vs. AI-first GTM org The structure flattens. Each individual contributor is amplified by a virtual team of agents. Fewer management layers, but each person is multiply effective. Figure 06 ### Org structure: traditional vs. AI-first GTM The traditional pyramid compresses. The IC's reach expands. Coaching and human development do not disappear. Span of control changes. Shane Paola Advisory21 09 · Org Structure Implications ## 9.3   What does not change In the excitement and anxiety around AI transformation, there is a real risk of overcorrection. Of assuming everything changes when the fundamentals remain constant. Worth being explicit about what AI does not replace: The mechanics of trust in complex B2B sales The importance of executive-level relationships and genuine multi-threading across buying committees The value of deep industry and customer knowledge built over time The role of human empathy in high-stakes buying decisions The leadership requirement to set direction, develop talent, and maintain culture AI changes the tools available to support these fundamentals. It does not replace them. Shane Paola Advisory22 10 · What This Is Not Section 10 # What This Is Not This framework is not about replacing people with AI. That framing is both wrong and counterproductive. The goal is supercharging your people with AI. Taking the same headcount and making it structurally more effective, more focused, and more capable than before. This is not a cost reduction play dressed as a transformation play. Efficiency gains are real, but they are a byproduct. The primary objective is competitive differentiation. The ability to out-execute on pipeline coverage, sales cycle velocity, and customer intimacy in ways that compound over time. This is not a tool-buying exercise. The risk of treating AI-first GTM as a technology procurement problem is that you end up with a more complex, more expensive stack and none of the cultural, process, or structural changes that make it effective. The tools are the last step, not the first. And this is not finished. Version 2.0 is a foundation. Every section will evolve as the technology matures, as early deployments produce real-world learnings, and as the best practitioners in this space develop the patterns that separate what works from what sounds good in a presentation. Not A people-replacement play The goal is supercharging. Same headcount, structurally more effective. Not A cost reduction exercise Efficiency is a byproduct. Competitive differentiation is the objective. Not A tool-buying exercise Tools are the last step, not the first. Foundations come before stack. Not Finished v2.0 is a foundation. The best patterns are still being written. Shane Paola Advisory23 11 · What Comes Next Section 11 # What Comes Next The next evolution of this document goes deeper into each function. Not the high-level framework, but the specific workflows, agent configurations, measurement approaches, and change management considerations for each GTM role. The intention is to build this into a genuine operating manual: something a CRO, VP Sales, or Head of GTM can actually use to run an AI-first transformation inside a real enterprise tech business. Functions to go deeper on in subsequent versions: Enterprise sales. Deal management, account planning, forecasting, and pipeline management workflows. SDR function. Outreach orchestration, pipeline qualification, and the human-to-AI handoff model. Channel and alliances. Partner segmentation, co-sell motion design, MDF allocation, and partner intelligence. Sales ops and rev ops. Data architecture, reporting infrastructure, and the agent ownership model. Customer success. Integration with the GTM motion and expansion revenue. Marketing and demand generation. In the AI-first context. Contact ## Shane Paola Advisory linkedin.com/in/shanepaola Version 2.0  ·  Living Document  ·  May 2026 Shane Paola Advisory24 =========================================================== # /ai-news =========================================================== # AI News — Shane Paola Weekly recaps and analysis of the AI stories that matter for B2B tech GTM leaders. Written by Shane Paola. ## Articles - [Jun 16–22, 2026 — Fable 5, Gemini 3, and the regulation reckoning](https://www.shanepaola.com/ai-news/ai-weekly-recap-jun-16-22-2026) — Markdown: [/content/ai-news/2026-06-22-ai-weekly-recap.md](https://www.shanepaola.com/content/ai-news/2026-06-22-ai-weekly-recap.md) - [Jun 1–8, 2026 — IPOs, autonomous attacks, and the coding transition](https://www.shanepaola.com/ai-news/ai-weekly-recap-jun-1-8-2026) — Markdown: [/content/ai-news/2026-06-08-ai-weekly-recap.md](https://www.shanepaola.com/content/ai-news/2026-06-08-ai-weekly-recap.md) =========================================================== # /ai-news/ai-weekly-recap-jun-16-22-2026 =========================================================== # AI Weekly Recap // Jun 16–22, 2026 # Government shutdowns, a $25B bond sale, and regulation's first fines **By Shane Paola · 22 June 2026 · 6 min read** Canonical: https://www.shanepaola.com/ai-news/ai-weekly-recap-jun-16-22-2026 The week the Fable 5 crisis hit Day 10 with no resolution, Google and OpenAI raced each other to GA, Nvidia rewrote the rules of AI finance, and the EU issued its first AI Act penalties. --- ## 01 · Safety / Policy — The Fable 5 affair: AI's first government shutdown The US government ordered Anthropic's flagship Fable 5 and Mythos 5 models offline on June 15. By Day 10, neither a deal nor a clear legal basis had materialised. What began as a disputed jailbreak escalated into a governance crisis with no precedent. Anthropic sent engineers to DC. They returned without a deal. Seventy-six cybersecurity experts signed an open letter demanding restoration. The White House pivoted from enforcement to a shared AI Security Severity Framework negotiation. Progress, but no timeline. By Friday, the NSA chief told Congress that Mythos had "breached almost all US classified systems in hours", a claim The Register then systematically unpacked, reporting the triggering "fix this code" instruction was never a jailbreak at all. The blowback was international and commercial. Europe was locked out with no recourse. The CEP called the ban "a geopolitical signal, not a security necessity." Macron raised it at the G7. Amazon was named by Fortune and WSJ as the firm that showed Commerce the original jailbreak. The Pentagon quietly confirmed it had replaced Claude with Grok in Project Maven, used to coordinate 2,000 Iran strike targets in 96 hours. Open source and DeepSeek moved into the enterprise gap. The episode has exposed a governance vacuum. No statutory basis, no due process, no hearing. A de facto AI kill-switch exists with no democratic guardrails around it. The Fable 5 ban did not end the week resolved. Sonnet 5 identifiers have surfaced in backend logs, suggesting Anthropic is already building past it. Sources: Shutdown announced (venturebeat.com); Fable Fiasco analysis (rstreet.org); Pentagon pivots to Grok for Maven (timesofindia.indiatimes.com); NSA Congressional testimony (ibtimes.co.uk); The Register: not a jailbreak (theregister.com); CEP: geopolitical signal (cep.eu). ## 02 · Model / Product — The model sprint: Gemini 3 Pro, GPT-5.6 and a week of releases Despite the Fable crisis, the frontier kept shipping. Google launched Gemini 3 Pro to all users on day one, GPT-5.6 hit 90% Polymarket odds with a 1.5M token context, and xAI topped the AI video leaderboard. Google moved fast and wide. Gemini 3.5 Flash became the global default in Search and the Gemini app mid-week, outperforming 3.1 Pro on coding and agents at lower cost. By Sunday, Gemini 3 Pro landed available to all users on day one with no waitlist. Google Cloud's London Summit added Gen-8 TPUs, Antigravity 2.0 and the Gemini Spark always-on agent. At Google I/O, Gemini received a Neural Expressive redesign alongside an Omni Flash video model. OpenAI's week was equally dense. GPT-5.5 Instant rolled to all 230 million free ChatGPT users, matching frontier thinking models on health queries. GPT-5.6 moved from stealth A/B test to near-certainty: 1.5M token context, an alignment fix and a 90% Polymarket launch probability by Monday, June 22. Anthropic's Claude Design shipped GitHub sync and bidirectional code integration with 1M+ enterprise users in week one. Nobel laureate John Jumper departed DeepMind for Anthropic the same week Noam Shazeer joined OpenAI. The talent moves of the year in a single week. xAI completed a strong week. Grok Imagine Video 1.5 topped the AI video leaderboard at $4.20/min versus Sora at $30/min, days before OpenAI announced Sora's September 24 sunset. Grok 4.3 launched on Amazon Bedrock with a 1M token context at $1.25/M. Microsoft added Claude Sonnet 4 and Opus 4.1 to M365 Copilot, and Grok embedded natively in Word, Excel and PowerPoint with a live X data feed. The first non-Microsoft AI inside Office. Sources: Gemini 3 Pro launch (thevergetoday); GPT-5.6 launch window (techtimes.com); GPT-5.5 Instant for 230M (digg.com); Grok Video 1.5 tops leaderboard (techtimes.com); DeepMind double departure (techcrunch.com); Grok embeds in Office (gagadget.com). ## 03 · Financial — Capital at scale: $7.6T forecasts and Nvidia's $25B bond sale Goldman Sachs published a $7.6 trillion cumulative AI capex forecast through 2031. Nvidia backed it up the same week with a surprise $25 billion bond sale, its first debt deal in five years. Goldman's "Tracking Trillions" report projected Nvidia capturing 75% of the $5.1T compute layer through 2031. Nvidia moved immediately. A $25B bond issuance funded $30B and $10B equity stakes in OpenAI and Anthropic respectively. Nvidia's Q1 FY2027 earnings confirmed the thesis. $38.6B in data center revenue with a Blackwell Ultra backlog exceeding $180B into 2027, plus the Vera CPU opening a $200B TAM for agentic AI infrastructure. Anthropic's Series H told a comparable story. $65B valuation at $47B ARR, up 5x in five months, heavily oversubscribed. Meta raised its annual capex guidance to $145B on Q1 revenue of $56B (+33%). Goldman projected a combined $4–5T market cap for Anthropic and OpenAI post-IPO. SpaceX debuted on public markets at +19%. Infrastructure plays filled the rest of the week. Baseten raised $1.5B at $13B, AI inference infrastructure led by Altimeter, Conviction and Spark. Odyssey AI raised $310M at $1.45B, notably choosing AWS Trainium over Nvidia for world model infrastructure. India's Sarvam AI became a unicorn at $1.5B with HCLTech taking a 10.5% stake. Sovereign AI momentum is now global. Sources: Goldman $7.6T forecast (goldmansachs.com); Nvidia $25B bond sale (ifre.com); Nvidia Q1 FY2027 (credencewire.com); Anthropic Series H (fortune.com); Baseten $1.5B (thenextweb.com); Sarvam AI unicorn (startupfortune.com). ## 04 · Policy / Regulation — Regulation's reckoning: EU AI Act enforcement begins The EU moved from legislation to enforcement this week. First fines issued, August 2 deadline confirmed, 80% of affected enterprises unprepared, and Macron called US export controls "strictly nationalist" at the G7. The EU AI Act Omnibus passed Parliament, delaying high-risk compliance to December 2027. Critics called it deregulation before the rules even apply. But the August 2 transparency and watermarking duties remain live, and enforcement has already started. The EU issued its first AI Act fines of €42M for biometric surveillance in public spaces. A separate EU Cloud and AI Development Act introduced a four-tier data sovereignty framework targeting a tripling of European data centre capacity. On preparedness, 80% of affected enterprises remain unprepared for August 2, facing fines of up to 7% of global revenue. The Five Eyes published a 23-category agentic AI security framework. 100+ best practices with the sobering note: "Assume it will behave unexpectedly." The NYSUT teachers' union voted unanimously to ban AI in schools through Grade 8. New York City's chancellor called it "the most invasive technology we've seen." The week's geopolitical flashpoint: at the G7, Macron called US AI export controls "strictly nationalist" and demanded democratic AI cooperation. A direct response to the Fable ban locking European operators out with no recourse. The Fable crisis is now formally an international trade and governance dispute, not just a US security question. Sources: EU AI Act Aug 2 (sentinel.ht); First AI Act fines €42M (credencewire.com); Five Eyes agentic AI guidance (mayerbrown.com); EU AI content transparency code (epium.com); Macron G7 (europeaninterest.eu); NYSUT school AI ban (chalkbeat.org). ## 05 · Model / Product — Platform wars: Big Tech absorbs the frontier Grok is now inside Office. Claude is inside Copilot. Nadella says OpenAI and Anthropic lack "societal permission", and Microsoft will build the governance layer. The integration race accelerated sharply. Grok embedded natively in Word, Excel and PowerPoint with a live X data feed. The first non-Microsoft AI model inside Office. Microsoft simultaneously added Claude Sonnet 4 and Opus 4.1 to M365 Copilot, and is evaluating DeepSeek V4 as a cheaper Copilot engine (Azure-hosted, national security tensions noted). Microsoft Copilot Cowork hit GA, agentic M365 at 50%+ of Fortune 500, running on Anthropic Opus 4.8 and Sonnet 4.6. Anthropic expanded Project Glasswing to 150 organisations across 15+ countries. 10,000+ critical-severity bugs found in month one. Claude Code Artifacts shipped live shared dashboards and interactive workspaces directly from the terminal. Mistral launched its Vibe enterprise agent platform with €4B committed to French and Swedish data centres, staking out Europe's sovereign AI infrastructure position. The talent dimension: Nobel laureate John Jumper left Google DeepMind for Anthropic, and Noam Shazeer joined OpenAI. Two of the biggest talent moves in AI history in the same week. Satya Nadella drew the strategic frame: OpenAI and Anthropic lack "societal permission", and Microsoft will build the governance layer between frontier labs and society. The platform consolidation thesis is now explicit. Sources: Grok in Office (gagadget.com); Claude Sonnet 4 + Opus 4.1 in M365 (thevergetoday); Copilot Cowork GA (irishtechnews.ie); Project Glasswing expands (anthropic.com); Mistral Vibe (thegreatentrepreneurs.com); Nadella societal permission (techtimes.com). --- ## Tools mentioned this week A few products I actually pay for and recommend. Links below are affiliate links. They cost you nothing extra and help support the writing. - **Wispr Flow** — Dictate anywhere on your Mac. Turns your voice into text in any app: Notion, Slack, email, code editors, wherever your cursor is, with near-zero latency. Sits in the background and activates with a hotkey. https://ref.wisprflow.ai/shane-paola - **Granola** — AI notepad for your meetings. Runs quietly in the background during any meeting (Zoom, Teams, Google Meet, in-person) and turns your rough notes into clean, structured summaries the moment you close the call. No bots, no recordings shared with others. Runs locally on your Mac. https://www.granola.ai?via=shane-paola - **Artefact** — Share AI-generated HTML privately. Upload the HTML, set a password and send a link. Guests open it in full fidelity, leave inline pins and sticky notes, you see every comment without handing over the source file or spinning up a server. https://useartefact.com --- ## FAQ **Why was Anthropic's Fable 5 taken offline?** The US government ordered Anthropic to take Fable 5 and Mythos 5 offline on June 15, 2026, following a disputed jailbreak incident. The government cited national security concerns, but critics including 76 cybersecurity experts and legal analysts at R Street noted there was no statutory basis, no formal hearing and no due process. The Register reported the triggering 'fix this code' instruction was not actually a jailbreak. As of June 22, Fable 5 remains offline at Day 10. **Has Google released Gemini 3?** Yes. Google launched Gemini 3 Pro on June 22, 2026, available to all users on day one with no waitlist. Earlier in the same week, Gemini 3.5 Flash became the global default in the Gemini app and Google Search AI Mode, outperforming Gemini 3.1 Pro on coding and agentic tasks. **When is GPT-5.6 releasing?** OpenAI's GPT-5.6 launch window opened June 22, 2026. As of that date, Polymarket placed the probability of a June 22–28 release at 90%. The model is expected to feature a 1.5 million token context window and an alignment fix identified in GPT-5.5. **What are the EU AI Act enforcement deadlines in 2026?** The EU AI Act's transparency and watermarking obligations take effect August 2, 2026, the same date as the EU's AI-generated content code. The EU has already issued its first fines (€42M for biometric surveillance in public spaces). High-risk AI system compliance has been delayed to December 2027 under the AI Act Omnibus passed in June 2026. An estimated 80% of affected enterprises remain unprepared for the August 2 deadline. **How much did Nvidia raise in its June 2026 bond sale?** Nvidia raised $25 billion in a surprise bond sale in June 2026, its first debt issuance in five years. The proceeds are being used to fund $30 billion and $10 billion equity stakes in OpenAI and Anthropic respectively. The same week, Nvidia reported Q1 FY2027 data centre revenue of $38.6 billion with a Blackwell Ultra backlog exceeding $180 billion. **Is Grok now integrated into Microsoft Office?** Yes. In June 2026, xAI's Grok embedded natively into Microsoft Word, Excel and PowerPoint with a live X (Twitter) data feed, making it the first non-Microsoft AI model to be integrated directly into the Office suite. Microsoft simultaneously added Anthropic's Claude Sonnet 4 and Opus 4.1 to Microsoft 365 Copilot. =========================================================== # /ai-news/ai-weekly-recap-jun-1-8-2026 =========================================================== # AI Weekly Recap // Jun 1–8, 2026 # IPOs, autonomous attacks, and the coding transition **By Shane Paola · 8 June 2026 · 5 min read** Canonical: https://www.shanepaola.com/ai-news/ai-weekly-recap-jun-1-8-2026 Five storylines that defined the most consequential week in AI so far. Anthropic and OpenAI both racing to Wall Street, and the first live AI cyberattack documented in the wild. --- ## 01 · Financial — The dual IPO sprint: Anthropic vs. OpenAI race to Wall Street Both Anthropic and OpenAI filed confidential IPO S-1s this week, targeting September listings at valuations exceeding $1 trillion each. Anthropic pulled ahead in the valuation race by closing a $65B Series H at $965B, overtaking OpenAI as the most valuable private AI company in history. The same week, it projected its first-ever operating profit of approximately $559M on $44B ARR, removing the last argument that frontier AI is an indefinitely loss-making enterprise. OpenAI countered by confirming its own confidential S-1 filing, with Goldman Sachs and Morgan Stanley leading a September listing at north of $1 trillion. DeepSeek also broke ranks from its no-outside-capital stance, raising its first external funding of $7B at a $50B valuation, signalling that the commercialisation phase is now a global race rather than a US duopoly. Alphabet moved in parallel, raising $80B in the largest US equity offering in history to fund AI compute buildout. Google also quietly signed a deal to pay SpaceX $920M per month for roughly 110,000 NVIDIA GPUs. The capital flows this week were unlike anything the tech industry has seen. Sources: Anthropic IPO S-1 Filing (anthropic.com/news/confidential-draft-s1-sec); $65B Series H (anthropic.com/news/series-h); OpenAI IPO Filing (cnbc.com); DeepSeek $7B raise (recodechinaai.com); Alphabet $80B raise (cnbc.com); Google–SpaceX $920M/month deal (techcrunch.com). ## 02 · Model / Product — The most crowded model launch week in AI history Every major frontier lab shipped simultaneously. Grok 4, Claude Opus 4.8, GPT-Rosalind, Nemotron 3 Ultra, MAI-Thinking-1 and Gemini inside Siri all landed in seven days. xAI's Grok 4 claimed new state-of-the-art on ARC-AGI V2 (15.9%) and Humanity's Last Exam (50.7%), with SuperGrok Heavy available at $300/month. Anthropic's Claude Opus 4.8 shipped with a 1M-token context window as default, 88.6% SWE-Bench score and dynamic workflow support, at unchanged $5/$25 pricing. OpenAI upgraded GPT-Rosalind with agentic coding and drug-discovery capabilities at 31% fewer tokens, and launched a GPT-5.5-Cyber variant for EU and ENISA. NVIDIA released Nemotron 3 Ultra with 550B open weights, the new US open-model frontier leader. Microsoft's MAI-Thinking-1 and six in-house models at Build 2026 formally began the company's decoupling from OpenAI dependence. The week's most symbolic moment: Apple WWDC crowned Gemini as the engine behind a rebuilt Siri, while Claude debuted as a built-in iPhone option for the first time. The two most consumer-visible AI surfaces on earth are now powered by Anthropic and Google, not OpenAI. Sources: Grok 4 (x.ai); Claude Opus 4.8 (anthropic.com); GPT-Rosalind (openai.com); Nemotron 3 Ultra (artificialanalysis.ai); Microsoft MAI-Thinking-1 (techtimes.com); Apple WWDC Gemini–Siri (tomsguide.com); Gemini 3.5 Pro 2M context (techtimes.com); GPT-5.6 leaks (chatforest.com). ## 03 · Safety / Incident — AI as weapon: autonomous cyberattacks arrive in the wild Three separate incidents confirmed this week that AI-powered attacks have crossed from theoretical to operational. The same tools defending infrastructure are now being used to breach it. Google caught the first AI-built zero-day exploit, staged for a mass 2FA bypass campaign, before it was deployed. Sysdig documented the first confirmed live LLM-agent intrusion: an attacker used an AI agent to pivot from a CVE to an internal AWS database in under an hour, across just four steps. Iran's APT42 was found using ChatGPT and Gemini to automate malware generation, phishing templates and influence operations at scale. On the measurement side, Cisco reported that multi-turn adversarial prompts bypass frontier model safeguards 39 to 55% of the time, a failure rate that renders current safety mitigations unreliable at scale. Character.AI and Google settled wrongful death cases. Grok faces deepfake lawsuits. The defensive counterpart: Anthropic's Project Glasswing deployed Claude Mythos to autonomously hunt vulnerabilities in critical infrastructure across 15+ countries, finding 10,000+ issues in month one, including a 17-year-old FreeBSD remote code execution flaw. The same capability that finds bugs can be turned to create them. Sources: Google AI Zero-Day (securityweek.com); Sysdig LLM Cyberattack (sysdig.com); Iran APT42 (techmeme.com); Cisco Safeguard Bypass Study (esecurityplanet.com); Project Glasswing (anthropic.com); AI Harm Litigation (bestlawfirms.com). ## 04 · Policy / Regulation — Washington vs. the states: a federal-state collision on AI law The US regulatory landscape fractured sharply this week, with federal and state governments moving in opposite directions simultaneously, and two major international deadlines locking in for August 2. The House introduced the bipartisan 'Great American AI Act', a bill that would freeze all state AI laws for three years, directly threatening Colorado's June 30 AI deadline and California's cascade of pending rules. The same week, Florida became the first US state to sue OpenAI, naming Sam Altman personally. It's a direct escalation of state-level enforcement power. Trump signed an executive order requiring AI companies to give the federal government early access to new models before public release. California's AI Transparency Act (SB 942) watermarking mandate locks in on August 2, requiring all major generative AI providers to watermark outputs. The EU's prohibited AI systems list takes full effect the same day, creating a dual deadline that every frontier lab must meet simultaneously. The EU also finalised its AI Act Omnibus deal: high-risk AI rules pushed to December 2027, with a new prohibition on AI-generated non-consensual intimate imagery added. Arizona proposed a 45% surcharge on data centre electricity to shield residential customers from the power costs of AI infrastructure. Sources: Great American AI Act Draft (obernolte.house.gov); Florida Sues OpenAI (cnbc.com); Trump AI Executive Order (cnbc.com); California AI Watermarks (gunder.com); EU AI Act Prohibited List (artificialintelligenceact.eu); EU AI Act Omnibus (consilium.europa.eu); Arizona Data Centre Surcharge (wsj.com). ## 05 · Research — Software engineering is being automated faster than anyone predicted Anthropic disclosed that Claude now writes more than 80% of its production code. The Stanford AI Index confirmed the shift is industry-wide. Developer jobs are down 20% in 12 months. The headline data point: Anthropic stated publicly that Claude authors more than 80% of all production code at the company, and may build its own successor model within two years. This isn't a demo or a benchmark. It's a running internal metric at a company deploying code at frontier scale. The Stanford AI Index 2026 contextualised the broader shift: agentic task success rates have tripled to 77% in 12 months, while developer job postings are down 20% year-on-year. Gen Z excitement about AI has fallen to 22%. People are starting to feel the labour market effects. Microsoft Build 2026 confirmed that coding infrastructure itself is being re-platformed: Project Polaris replaces GPT-4 in GitHub Copilot, Windows becomes an Agent Runtime, and MAI-Thinking-1 runs without OpenAI data. GitHub Copilot's switch to AI Credits billing signals that usage-based metering, not flat seats, will define the economics of AI-assisted development going forward. Sources: Anthropic Claude 80% (venturebeat.com); Stanford AI Index 2026 (hai.stanford.edu); Microsoft Build 2026 (chatforest.com); Copilot Credits Billing (github.blog); Karpathy Joins Anthropic (techcrunch.com). --- ## Tools mentioned this week A few products I actually pay for and recommend. Links below are affiliate links. They cost you nothing extra and help support the writing. - **Wispr Flow** — Voice-to-text, everywhere. Turns your voice into clean, formatted text in any app: Slack, email, docs, your IDE. The fastest way I've found to draft long-form thinking without losing the thread. https://ref.wisprflow.ai/shane-paola - **Granola** — AI notes for real meetings. Listens to your calls and turns your scratch notes into structured summaries, action items and follow-ups. Runs locally on your Mac and is the only meeting tool I've stuck with. https://www.granola.ai?via=shane-paola - **Artefact** — Share AI artefacts privately. Upload an AI-generated HTML report, dashboard or prototype, set a password and send a link. Guests open it in full fidelity, leave inline pins and sticky notes, you see every comment without handing over the source file. https://useartefact.com --- ## FAQ **Did Anthropic file for IPO in 2026?** Yes. Anthropic filed a confidential IPO S-1 with the SEC in June 2026, targeting a September Wall Street listing alongside OpenAI. The same week, Anthropic closed a $65B Series H at a $965B valuation, the highest ever for a private AI company, and projected its first-ever operating profit of approximately $559M on $44B annualised revenue. **What is Claude Opus 4.8?** Claude Opus 4.8 is Anthropic's flagship model, released on June 8, 2026. It features a 1 million token context window by default, scores 88.6% on SWE-Bench (a software engineering benchmark) and supports dynamic agentic workflows, all at the same $5 input / $25 output per million token pricing as its predecessor. **Has an AI model been used in a real cyberattack?** Yes. In June 2026, Sysdig documented the first confirmed live LLM-agent cyberattack, in which an attacker used an AI agent to pivot from a CVE to an internal AWS database in under an hour across 4 steps. Separately, Google intercepted the first AI-built zero-day exploit before deployment, and Iran's APT42 was confirmed to be using ChatGPT and Gemini to automate malware and phishing at scale. **What is the Great American AI Act?** The Great American AI Act is a bipartisan House discussion draft introduced in June 2026. It would impose a 3-year federal moratorium on all state AI laws, freezing California's AI Transparency Act, Colorado's pending AI rules and similar state-level legislation, while a federal framework is developed. **What percentage of code does Claude write at Anthropic?** As of June 2026, Anthropic disclosed that Claude authors more than 80% of all production code at the company. Anthropic also stated that Claude may build its own successor model within two years, the first time a frontier lab has publicly projected AI-authored AI at that level. **What did Apple announce at WWDC 2026 about AI?** At WWDC 2026, Apple revealed that Google's Gemini now powers a rebuilt Siri, replacing OpenAI. Apple also debuted a new AI Extensions system that allows Claude to be installed as a built-in iPhone AI option, the first time Anthropic's model has had native iOS integration.