The Board Room

Deep dives

1. 01

   The AI Workforce Paradox: Data Says Hire More, Quality Says You're Building a Mediocrity Machine

   The Paradox in Two Data Points

   The Ramp/Revelio Labs study across 21,000+ US firms just produced the most significant empirical challenge to the 'AI destroys jobs' thesis: heavy AI adopters grew headcount 10% over two years, with entry-level roles growing even faster at 12%. The firms winning aren't automating humans away — they're discovering that AI creates demand expansion (falling cost per task makes more tasks viable) and generates new supervision work that didn't exist before.

   But a parallel analysis names the hidden cost: 'synthetic seniority.' AI-assisted work converges at a quality level that is competent but undifferentiated — roughly 70% of what an expert would produce. High enough to clear review. Low enough to lose any competition for attention or loyalty. The Coca-Cola AI Christmas ad is the case study: technically proficient, dramatically cheaper, received as soulless.

   Deploy AI across every function, cut headcount to match, and within 18-24 months a quality ceiling gets engineered into the organization that is both invisible and irreversible.

   The Compounding Trap

   The danger isn't just that output quality converges at 70%. It's that the 'expert eye' — the senior ICs who can distinguish 70% from 100% — are the same people being reduced through efficiency-driven restructuring. When they leave, the 70% standard becomes the new 100%. An organization can forget what its best looked like within a single leadership generation.

   The 'never skilling' problem compounds this further. When AI handles the cognitive work that traditionally built professional judgment, you create a generation of workers who are productive with AI but helpless without it — and critically, who lack the judgment to know when AI is wrong. The entry-level hiring surge the data recommends (12% growth) may be building a pipeline of workers who never develop the discrimination skills needed for senior leadership.

   The Resolution: Human-Machine-Human Architecture

   The answer is not to stop hiring or stop deploying AI. The answer is a specific workflow architecture:

   1. Human judgment opens — the taste call, the strategic frame, the 'what great looks like' that no brief specifies
   2. AI accelerates the middle — drafting, variants, iteration, testing at machine speed
   3. Human judgment closes — quality gate, the final call, the refinement that moves 70% to 100%

   Organizations that invert this order — letting AI set the opening frame or make the final decision — permanently cap their output at the tool ceiling. The strategic question: do you still employ people who know what great looks like, and do they have the authority to demand it?

   The Board Narrative Needs Reframing

   If your AI investment thesis is primarily a cost-reduction story, the data says you're leaving the larger prize on the table. The 10% headcount growth firms aren't spending less — they're capturing new market opportunities that pure-automation players can't reach because they lack the human judgment layer. The board framing should shift from 'efficiency/cost reduction' to 'growth acceleration via capability expansion' — but with an explicit quality governance layer that prevents the 70% ceiling from becoming structural.

   What to do

   1. →

      Identify your 'expert eye' concentration risk this quarter — map which senior ICs are the only people who can distinguish 70% from 100% in their domain, and flag them as critical retention targets

      This sprintOnce these people leave through efficiency-driven reductions, the quality standard is irreversibly lowered — and you won't know it happened until customers notice
   2. →

      Implement Human-Machine-Human workflow architecture as doctrine for all customer-facing and product work by end of Q3

      This quarterThe empirical data shows this is where the 10% growth companies are investing — they're redesigning decision loops, not just automating tasks
   3. →

      Redesign performance evaluation to test judgment quality, not output artifacts, starting next review cycle

      This quarterCurrent evaluation systems can't detect synthetic seniority because they measure what was produced, not whether human discrimination was applied
   4. →

      Design deliberate skill-building programs for entry-level AI-native hires that create 'AI-off' periods for developing domain judgment

      This quarterThe 12% entry-level growth creates pipeline risk if these hires never develop the judgment to know when AI is wrong
2. 02

   Hybrid Inference Just Crossed the Tipping Point: 71% of Your Cloud API Spend Is Waste

   The Stanford Numbers That Change the Math

   Stanford's latest study quantifies what many suspected: 71.3% of queries currently sent to cloud LLMs can be handled by local models — up from 23.2% in 2023. With intelligent routing across model variants, coverage reaches 88.7%. The realized savings: 59% cost reduction and 64% energy reduction. The 5.3x improvement in intelligence-per-watt over two years isn't incremental — it's exponential compression of the cloud LLM's addressable market.

   If you're paying frontier prices for commodity inference, you're subsidizing your vendor's margin on work that a $2,000 GPU can handle.

   AMD's Framework Advantage Makes Dual-Sourcing a No-Brainer

   The AMD/Wafer benchmark result amplifies the architecture shift. Serving GLM-5.2 at 2x lower cost than NVIDIA Blackwell — achieved through sglang selection, MXFP4 quantization, and configuration tuning rather than proprietary silicon — means this is a repeatable methodology, not a one-off win. Combined with Qwen 3.6 27B running at 32 tok/s on consumer hardware with production-quality output, local inference has crossed from experiment to enterprise-viable.

   The MoE Architecture Unlock

   The Mixture-of-Experts architecture (exemplified by multiple open models this week at 1.6T total parameters but only 48B active) makes trillion-scale models economically viable for self-hosting by mid-size engineering organizations. The cost curve for hosting frontier-equivalent capability on-premises is falling faster than cloud API pricing can adjust.

   Metric	2023	2025	Change
   Local model coverage	23.2%	71.3%	+207%
   With intelligent routing	~40%	88.7%	+122%
   Intelligence per watt	1x baseline	5.3x	+430%
   AMD vs Nvidia cost	Parity	2x advantage	50% savings

   What This Means for Your Architecture

   The cloud LLM API business model is being squeezed into ~11% of queries that genuinely require frontier reasoning capability. The strategic response isn't 'go fully local' — it's intelligent routing that sends commodity queries to local/edge inference and reserves expensive frontier API calls for the tasks that genuinely require them. This is the infrastructure equivalent of right-sizing your compute: most organizations are paying frontier prices for commodity work because they never built the routing layer.

   For any executive planning GPU procurement, the negotiating dynamic with NVIDIA just changed. Dual-sourcing isn't just supply chain prudence — it's a 50%+ cost optimization opportunity that's repeatable across production workloads today.

   What to do

   1. →

      Commission a hybrid inference architecture assessment within 30 days — model current LLM API spend, identify commodity query percentage, calculate ROI of local-cloud routing

      NowStanford data proves 59% cost reduction is achievable today with existing tools; every month of delay is wasted spend
   2. →

      Initiate AMD MI355X evaluation for inference workloads — run parallel benchmarks against NVIDIA fleet on production traffic patterns by end of Q3

      This sprint2x cost advantage through framework optimization alone makes dual-sourcing an immediate savings opportunity, not a future hedge
   3. →

      Evaluate MoE-architecture open models (LongCat-2.0, GLM-5.2) against current API spend on coding and general-purpose workloads

      This quarter48B active parameters at 1.6T total means self-hosted frontier-equivalent is viable for mid-size orgs; the economics favor testing now
3. 03

   AI Security at Machine Speed: Your Remediation Capacity Is Now the Binding Constraint

   The Discovery-Remediation Gap Has Broken Open

   Three converging signals this week confirm that AI has shattered the security equilibrium:

   • Claude Mythos Preview triggered a 3.5x monthly spike in high/critical CVE disclosures — a single model release fundamentally altered the attack surface
   • The Linux Foundation's Akrites launch confirms mean-time-to-exploit is now 'negative days' — exploits precede patches in open-source ecosystems
   • Anthropic's Fable 5 was offline for 19 days due to government intervention triggered by a jailbreak vulnerability — regulatory takedowns are now an operational assumption, not an edge case

   The board-level question is stark: has your security team's remediation capacity grown 3.5x? If not, your effective exposure has.

   The New Risk Category: Regulatory Takedowns as Downtime

   The Fable 5 incident introduces regulatory intervention as a measurable operational risk. A frontier model went dark for 19 days not due to technical failure but government action triggered by Amazon researchers discovering a jailbreak. Anthropic's response — a classifier achieving >99% catch rate — is technically elegant but strategically revealing: even model providers now architect for regulatory takedown as a design assumption.

   China's Z.ai has turned this into competitive positioning. GLM-5.2 — 744B parameters, MIT-licensed, trained on Huawei silicon — is explicitly marketed as 'the model that can't be taken away from you.' Every US regulatory takedown strengthens this value proposition for non-US customers. Your international customers now have a credible alternative that exploits your regulatory vulnerability.

   The Industry Response: Formalizing AI Security as Compliance

   The industry is drafting CVSS-like severity scoring for jailbreaks (Anthropic, Amazon, Microsoft, Google participating). Both Anthropic (Glasswing) and OpenAI (Daybreak) are building parallel vulnerability discovery programs that will further accelerate the CVE discovery rate. This means the gap between discovery speed and remediation speed will widen before it narrows.

   The architectural response must go beyond faster patching. It requires: reduced blast radius through model segmentation, zero-trust supply chain verification per the Linux Foundation's framework, and multi-model failover that treats regulatory takedowns as equivalent to infrastructure outages. The reactive vulnerability management model is broken at machine-speed discovery rates.

   What to do

   1. →

      Elevate AI-driven vulnerability remediation to board-level risk discussion at next board meeting — present the 3.5x discovery rate increase against current patch SLAs

      This sprintDiscovery at machine speed with remediation at human speed means effective exposure is growing monthly; this is a governance gap, not just an ops gap
   2. →

      Conduct a vendor concentration risk audit specifically stress-testing for regulatory takedown scenarios — model the impact of a 19-day outage of your primary model provider

      This sprintFable 5 proved this is not theoretical; any model provider is one jailbreak disclosure away from government-mandated downtime
   3. →

      Stand up or formalize an AI security practice aligned with emerging CVSS-for-jailbreaks framework by end of Q3

      This quarterThe framework is being drafted by Anthropic/Amazon/Microsoft/Google now — early adopters will shape standards; laggards will be governed by them
   4. →

      Build multi-model failover architecture that treats regulatory takedowns as equivalent to infrastructure outages

      This quarterZ.ai's 'no kill switch' positioning proves competitors will exploit your regulatory risk; your architecture must be resilient to it
4. 04

   The PagerDuty Death Pattern: How AI Kills Single-Function SaaS — and How to Audit Your Own Exposure

   The Pattern: Feature Absorption at AI Speed

   PagerDuty's loss of Uber after 12+ years isn't an anecdote — it's a structural pattern now accelerated by AI. A tech commentator's tweet about leaving PagerDuty received 610,000 views with overwhelmingly confirming responses. This is a preference cascade — the moat didn't erode gradually; it collapsed when adjacent platforms (Datadog for monitoring-native teams, Sentry for developer-native teams) added incident management as a feature rather than a product. AI compressed the development cost of 'good enough' functionality to near zero.

   The Confirmation Signal: Talent Exodus Precedes Revenue Loss

   Five9 provides the leading indicator. Labeled an 'AI loser' in September 2024, twenty months later its CRO, VP of Product Engineering, and CTO have all departed within weeks. The internal talent knew the strategic math before the market priced it. When your best people leave simultaneously, they're not disagreeing with each other — they're agreeing about the future.

   Meanwhile, Alibaba's PageAgent reduces in-product AI agent capabilities to a single script tag. Any SaaS product whose AI copilot was a key differentiator six months ago is now competing against free, embeddable alternatives that any developer can integrate in hours.

   The question every technology executive must ask: which of my products exists as a standalone because the adjacent platform hasn't bothered to replicate it yet? AI changes the economics of 'bothering.'

   The VantageScore Lesson: Moats Collapse Instantly When Switching Costs Are Removed

   VantageScore grew from 3% to 10% penetration at UWMC in a single month once Fannie/Freddie accepted it in securitization data. FICO's dominance was never purely product superiority — it was regulatory mandate and integration lock-in. The moment structural barriers fell, adoption went exponential. In the AI era, structural lock-in is being systematically dismantled — by regulators, by open-source alternatives, and by platforms that abstract away switching costs.

   The Audit Framework

   Evaluate every product in your portfolio across three dimensions:

   1. Is the core function replicable as a feature of an adjacent platform? If yes, you're on a timeline.
   2. Is your moat genuine product differentiation or merely switching-cost/regulatory lock-in? Lock-in is evaporating across every sector.
   3. Would AI reduce the development cost for a competitor to replicate your core function to near zero? If yes, assume they will — the question is when, not whether.

   What to do

   1. →

      Conduct a 'PagerDuty risk audit' this quarter — identify which products in your portfolio could be replicated as features of adjacent platforms with AI acceleration

      This sprintThe pattern shows moats collapse in a preference cascade, not gradually — by the time it's visible in revenue, it's too late to respond
   2. →

      Monitor executive departures at competitors and acquisition targets — coordinate with recruiting to approach Five9's CRO, VP Engineering, and CTO who are all recently available

      NowTalent signals precede business signals by 12-20 months; these executives have deep contact center AI experience and strategic visibility into the displacement pattern
   3. →

      Evaluate whether your competitive moats are genuine product differentiation vs. structural lock-in — model what happens if switching costs are eliminated by open-source or platform abstraction

      This quarterVantageScore's 3%→10% in one month proves monopoly positions collapse overnight when barriers are removed; your planning needs to account for this scenario