FlashFeed

Full text

There's an important distinction between AI that just works today, and AI that lasts at scale. Many companies optimize hard for the first one without ever asking whether they're building the second.

Velocity without discipline and strategic direction is a liability, not an asset. The hardest part of building AI at scale isn't getting a model to work once. It's building systems that continue to work, scale beyond individual teams and use cases, and improve consistently over time.

Today's AI systems do more than just predict and optimize. They converse, reason, and increasingly take action. An autonomous system making decisions on a traveler's behalf creates a very different set of expectations around reliability, governance, and accountability. As AI takes on more of those roles, the principles behind how these systems operate matter more than ever.

We have spent years applying AI and machine learning (ML) across the traveler journey — from personalization, ranking, and recommendations, to fraud prevention, customer support, and, more recently, generative and agentic AI experiences. That depth of experience is what led us to develop a set of ML and AI principles to guide how we build, deploy, and evolve AI systems across our company.

The goal is simple: Make sure the systems we build create real business value, scale, and operate safely. These principles define how we measure, design, govern, and operate our systems.

From principles to practice

Publishing principles is the easy part. The harder and more important work is turning them into operating mechanisms: Recommendations, requirements, tooling, and release processes that teams actually use.

We have begun using 'Agentic Release' tollgates: A set of recommended and, in some cases, required checks before launching agentic AI features. These tollgates translate principles like clear ownership, risk-based governance, evaluation, safe rollout, and monitoring into concrete expectations for teams.

Some of these recommendations and requirements are already being automated and integrated into the software development lifecycle (SDLC). Over time, the goal is for these expectations to become embedded in how we design, evaluate, approve, launch, and monitor AI systems from the start.

Outcomes: Measuring what actually matters

The first test for any model is whether it improves a business outcome and, ultimately, the traveler experience — not whether it just improves a technical metric.

Align models to metrics with business impact: Every ML effort must tie directly to a key business outcome or traveler experience metric. Technical optimizations are useful midpoints, not end goals .

Optimize for return on cost : The value a model creates has to justify what it costs to develop, train, and monitor, plus the operational complexity it adds. Favor solutions that deliver lasting impact relative to what they cost to run.

Justify complexity against strong baselines: Complexity should be earned, not assumed. Start with a strong baseline: An existing general model, a simple heuristic, an off-the-shelf solution. Reach for specialized models or more complex architectures only when simpler options genuinely can't meet the bar.

Require both offline and online evaluation : No model goes to broad deployment on offline validation alone or jumps straight to A/B testing. Every model must perform in both offline and online evaluations. Over time, our offline evaluations should reliably predict what we see online.

Design: building systems that scale beyond the teams that build them

Getting a model to work is one challenge. Making its value extend beyond a single team or use case is the harder one.

Build on shared foundations; specialize only when justified: Favor shared, platform-wide foundations for core capabilities, data representations, and model building blocks. Specialization should build on those foundations, not spin up isolated stacks, so when the foundation improves, the gains flow across the organization.

Treat data as a first-class product : A model's quality is bounded by the quality of its data. We need to maintain robust pipelines, clear lineage, reproducibility, and reusable features built with documented ownership, clear schemas, and SLAs that other teams can rely on.

Prioritize generality over local optimization : When two approaches perform similarly, favor the one whose learnings, assets, and operating patterns can be reused across teams, brands, and use cases. We should optimize not just for local performance, but for how quickly improvements can diffuse across the company and compound over time.

Minimize and sunset manual business rules: Manual rules are sometimes necessary for policy, safety, or compliance, but they should be explicit and reviewed regularly, never silent patches for weak models or a source of permanent maintenance debt.

Reproducibility and traceability by default : Training data, features, configurations, evaluation results, deployment versions, and key decisions should all be documented and recoverable. That's what lets you debug a production issue months later and hand off ownership without losing institutional knowledge.

Trust: ownership, governance, and operating responsibly at scale

The bar for deploying AI isn't just "does it work?" It's "can we stand behind it?" Trust isn't something you add at the end; it's earned over time and maintained across the full lifecycle of every model we ship.

Assign clear ownership and accountability: Every model needs defined ownership across its lifecycle — a business owner, a product owner, an AI owner, and an operational owner. These don't need to be four people, but the responsibilities must be explicit. Who's accountable for outcomes? Who responds if the model drifts? Who answers the incident at 2 a.m.? Without this in place, models become orphaned and problems surface with no one to own them.

Adhere to standards and governance: AI and ML models must use approved platforms and comply with established company standards, release gates, and governance processes. Operating outside these guardrails requires a clear, defined path to remediation or deprecation, rather than an open-ended exception.

Govern proportionally to risk : The level of review, evaluation rigor, and human oversight should scale with a model's impact. A customer-facing model that affects pricing or availability for millions of travelers demands a far higher bar than an internal tool used by a small team. For high-impact, safety-sensitive, or highly autonomous systems, human-in-the-loop checkpoints are built in from the start.

Design for fairness, privacy, and transparency : We actively test for unintended bias, have strong data guardrails, and favor explainability when decisions meaningfully affect users. These are incorporated from the start, not added on.

Design for safe rollout, rollback, and control : Deployments are progressive, with rollback paths, fallback mechanisms, and circuit breakers ready before launch. The ability to safely undo a deployment matters as much as the ability to ship it.

Monitor continuously and adapt: Once live, teams must actively monitor quality, drift, latency, cost, and business performance and retrain or recalibrate when the data shifts. A team should always be able to explain how its model is performing now, not just how it performed when it launched.

These principles do more than define how we build. They define what we're willing to ship and how we stand behind it. In a world where AI systems are increasingly consequential and make real decisions for real travelers and partners, these standards matter. Applied consistently, they build responsible AI that lasts.

Xavi Amatriain is Chief AI and Data Officer at Expedia Group

Xavier will share more details about Expedia's architecture during his session at VB Transform on July 14 at 11:10 am PT. He will discuss: "Expedia's blueprint for building autonomous agents for high-stakes transactional systems."

Interested in attending VB Transform 2026? Register here . A select number of complimentary passes are also available to senior technology leaders. Contact us to get yours.

Full text

A secret shopper study has pulled back the veil on the practices of nearly 50 telehealth sites that prescribe popular GLP-1 medications for weight loss.

Direct-to-consumer telehealth companies have grown rapidly in response to patients’ demand for GLP-1s like semaglutide and tirzepatide, especially after shortages created a market for compounded versions that aren’t approved by the Food and Drug Administration. As online prescribing has grown, clinicians and public health experts have raised concerns about the potential for lax virtual care practices to put patients at risk.

In a study published Monday in the Journal of the American Medical Association, one researcher at Yale University set out to document those online prescriptions by posing as a patient. The results capture what many Americans have learned in the last few years: It is extraordinarily fast and easy to get a GLP-1 online.

Continue to STAT+ to read the full story…

Full text

England, a country built on brick designed to retain heat, experienced a historic heatwave last month. Having experienced the blistering 37°C/98.6°F peak myself, there's little I would not have done to escape that heat—even donning a silly-looking personal cooling wearable didn't seem beyond the pail.

Hardware channel Making Stuff with Mike is based in Denmark, which has also experienced its share of record-breaking heat . So, the YouTuber created a "functioning art piece" to keep him cool during the in-tents heat of BornHack 2026 . The 'art piece' in question is a hard hat with an unusual centerpiece jutting out of it: A mini-chimney with a 120 mm Noctua fan mounted on top (via Hackaday ).

The fan structure is 3D printed, with a custom USB-to-fan-header adapter inside. The fan itself runs off of 5000 Ah power banks, which Mike hopes will keep him cool for about six hours at a time. Airflow from the Noctua fan to Mike's bonce passes through a drilled out hole in the top of the hard hat, though this prototype's fan may still struggle when temperatures rise again.

The current fan setup isn't very powerful, running at five volts. Mike says he'd like to eventually upgrade to a 12 volt design for better airflow, though his chief concern is that this would drain the power bank much faster and require a physically larger unit that may not fit inside the chimney structure.

Mike also admits he's not thought much about cable management, electing to simply duct-tape the power bank to the back of the helmet for now. Though very silly-looking, it's hard not to be charmed by this particular project or its creator.

For example, when mounting the chimney template to the hard hat, Mike inspects his work and quips, "It's not perfectly straight but, you know what? Neither am I, so I don't care!"

Though an early prototype, Mike already has plans for how he'll further iterate on the DIY cooling device. Even at this early prototype stage, the 3D printed chimney features a spot to mount a PCB internally and potentially facilitate a light-up design down the road. Mike is also weighing up whether to leave the (now unsafe) safety hat bright yellow, or give it a steampunk makeover.

Full text

The news was almost predictable. For yet another year, auditors have found that America’s new weapons aren’t being delivered on time.

“ DOD plans to invest over $2.4 trillion to develop and acquire its costliest weapon programs,” according to the Government Accountability Office’s 2026 annual report on major defense acquisition programs, or MDAP.

“Schedule delays persisted across MDAPs, signaling overly optimistic time frames,” according to the GAO report, which examined 104 of the Pentagon’s most expensive weapons programs . “The overall average time frame to deliver a capability increased this year to over 12 years. Further, several MDAPs have not set new delivery dates or are delaying critical interim milestones.

“By keeping delivery dates static, these programs raise questions about how realistic their estimates are . This means the 12-year average will likely increase in the future.”

If that sounds familiar, the 2025 annual report found that the DOD “continues to struggle with delivering innovative technologies quickly and within budget.”

However, the 2025 study seemed particularly concerned by rising costs of weapons due to inflation in the U.S. economy. In that regard, the 2026 analysis revealed a mixed price picture, with 46 of 72 programs reporting increases totaling $122 billion, and 16 reporting decreases totaling $47 billion.

GAO’s 2026 report seems more focused on program delays, especially in middle tier of acquisition , or MTA, projects that are intended to quickly jumpstart fielding of weapons within five years. The Pentagon is investing more than $49 billion across 23 of the most expensive MTA projects.

But GAO found that many systems are being slotted into the MTA despite the technology being too immature to be fast-tracked.

“For example, between 2018 and 2025, 18 out of 40 programs have entered the MTA pathway with immature technologies,” the report stated.

Many were lower than Technology Readiness Level 6 (prototype stage), and some were less than TRL 3 (proof of concept stage). Of eight current MTA projects, GAO deemed seven to be technologically immature, including the Hypersonic Attack Cruise Missile and the Next Generation Persistent Overhead Infrared sensor.

Thus, “programs are increasingly using the MTA pathway to mature technologies, when the intention of the pathway is to prototype and or field a residual capability within two to five years,” Shelby Oakley, GAO’s director for contracting and national security acquisitions, told Defense News.

“This is why we are not seeing capabilities fielded any faster,” Oakley said.

Among specific program issues, the Navy’s MQ-25 Stingray carrier-based unmanned aircraft has a 2.5-year delay to “initial operational capability and a 26-month delay to the end of initial operational testing,” GAO found. Changes to the design based on testing could push deployment out even longer.

For the DDG(X) destroyer, the Navy has yet to offer an acquisition strategy that details the assumptions behind the project.

“As such, the Navy’s business case for the DDG(X) program is not apparent,” GAO said. The DDG(X) is also likely to be affected by shipyard delays in building the Arleigh Burke-class Flight III series, which are up to 55 months behind schedule.

The Air Force’s B-52 radar modernization program, meanwhile, is beset by “cost increases and schedule delays.” The Air Force is opting to enter production with “very little development flight testing completed.”

The Army’s Mid-Range Capability program, which uses mounts Navy missiles and vertical launch tubes in Army vehicles, has “identified three additional critical technologies, all of which are immature.”

There are questions on “whether the return on investment is worthwhile for the four batteries planned for production by the Army,” the GAO stated.