Ex-Lululemon Executive Says AI Adoption Is Slowing

Why is the A.I. boom feeling stalled in the C-suite? Julie Averill, a former Lululemon executive, argues the answer isn't weak models but the expensive, slow plumbing companies refuse to pay for. In a guest essay published on Aug. 3, 2026, Averill writes that “the A.I. revolution is stalling because companies don’t want to admit that integrating the technology is expensive and slow and requires human effort,” and urged leaders to stop pretending the hard work is optional.https://www.nytimes.com/2026/08/03/opinion/ai-hype-tech-layoffs.html

Why this argument lands: the claim reframes current pain points from model performance to organisational readiness. Averill—who the New York Times identifies as a former senior executive and whose profile credits her with supporting Lululemon’s growth during a multi-billion-dollar expansionhttps://www.nytimes.com/by/julie-averill—says the technology is now the easy part and that data quality, governance, workflow readiness and human capability are the real constraints. On a recent podcast she summed the state of retail AI as “Maybe the first inning.”https://podcasts.apple.com/us/podcast/ais-first-inning-julie-averill-former-lululemon-rei/id1434573070?i=1000776240818

The hidden bill for AI: messy data and change management

Averill’s central point is mundane but consequential: training and deploying models is only a fraction of the work. Firms that rush pilots without fixing siloed product catalogs, inconsistent master data, and fragmented decision processes see limited lift. Retail-focused reporting already shows examples where Lululemon and peers pilot personalization or demand-forecast systems but then run into integration drag across merchandising, supply chain and stores.https://www.digitalcommerce360.com/2026/07/30/how-lululemon-is-using-ai/

That diagnosis tracks with consultants who have watched enterprise IT projects for decades: the bulk of cost sits in data cleansing, workflow redesign and governance, not in the model itself. Averill’s contribution is bluntness — she calls out executives who treat proof-of-concept metrics as business outcomes.

Why now: hype, layoffs and a demand for return on investment

Averill published her piece amid a wave of tech-sector scrutiny — firms cutting staff, boards demanding clearer ROI on AI bets and boards asking what’s actually changing in operations. Critics outside the essay warn that an attention-grabbing narrative can obscure nuance: one industry commentator compiled a list of overblown AI headlines to caution readers against mistaking PR for progress.https://www.fool.com/investing/2026/07/12/the-ai-headlines-you-should-be-most-suspicious-of/

That sceptic’s point matters because many vendors and start-ups sell “quick wins” that depend on labor-intensive configuration and ongoing human review. Averill explicitly urges leaders to budget for that labor — and to build the governance that prevents models from producing brittle or misleading recommendations.

What critics might say about Averill’s framing

Averill’s essay is persuasive to leaders familiar with operations; it risks underplaying genuine technical gaps, especially in safety-critical domains where models still make avoidable errors. The broader debate needs more empirical measurement of where pilots fail: is it data, poor integration, unrealistic expectations, or a mix of all three? Industry summaries that amplify cautionary headlines highlight the question but don’t yet supply systematic evidence beyond corporate anecdotes.https://aisight.fractal.ai/select-reads/leadership-beyond-technology-ai-transformation-future-of-work/

Averill’s prescription — treat AI adoption as a transformation, not a plugin — will ring true for chief operating officers and technology officers who have seen model-centric pilots fizzle. Vendors, by contrast, have commercial reason to keep marketing the easy story.

If she’s right, the near-term winners won’t be the firms that bought the flashiest models but those that invest in data engineering teams, clearer decision rights and governance processes. Boards that once cheered a model demo will now have to approve multi-year investments in slow, unglamorous work. That shift will be the test of whether A.I. moves from pilot to predictable productivity.