AI and Software Engineering

From Hype to Scale: Operationalizing AI in Retail Enterprises

Artificial intelligence is reshaping retail at every layer of the enterprise, from merchandising and supply chain optimization to customer experience and workforce productivity. Yet many organizations struggle to move beyond isolated pilots and disconnected proofs of concept.

In this session, Keith Henning explores the realities of enterprise AI adoption inside large retail environments, focusing on the organizational, architectural, and operational changes required to successfully scale AI initiatives. Drawing from real-world experience leading enterprise technology strategy, Keith will discuss how retailers can align AI investments with measurable business outcomes while navigating governance, legacy systems, data complexity, and organizational resistance.

Attendees will gain practical insights into building AI-ready platforms, enabling cross-functional adoption, identifying high-value use cases, and creating sustainable innovation models that move beyond experimentation into enterprise-wide transformation.

From Exposed to Protected: Securing Enterprise LLM Applications with SapienShield AI Firewall

A 30-minute practitioner session for AI in Software Engineering (SoftEd), September 24, 2026 — walking engineering leaders through real AI threat scenarios and demonstrating how a firewall architecture addresses each risk without rebuilding existing application stacks.

AI hasn't killed authenticity (yet) - but it's thrown the definition up for grabs.

In a world where anyone can generate anything in seconds, the question isn't whether you used AI. It's whether you still showed up. Shannon McGarity and Elizabeth Zietlow of Hover Media, bring a practitioner's lens to the messiest question in AI-augmented work right now: how do you stay you — in your outputs, your brand, and your working relationships — when the tools can do so much of the heavy lifting? Drawing on their own experiments, missteps, and hard-won working principles, they'll explore what authenticity actually means when the AI badge lands on work you spent nine months thinking about, why your personal values around AI use matter as much as the brand you work for and how articulating your own "AI Credo" — your "this is how I show up" — might be the most underrated professional move of the moment.

Overview of Strategies for Managing OpEx for Individual and Workflow AI Usage with Open Weight Models

This session will walk through several real-world examples of replacing OpenAI/Anthropic model usage with open weight-based models to help stabilize cost profiles, improve usage predictability, and strengthen data access controls.

99% Confidence, 100% Wrong When AI gets bold, QA gets busy.

Artificial intelligence was supposed to make life easier. Faster answers, smarter tools, fewer mistakes. Then it showed up to work with the confidence of a senior architect, the accuracy of a broken autocomplete, and the emotional stability of a production server on release day.

This session is a hilarious look at the strange new partnership between QA engineers and AI: one confidently generates answers, the other quietly asks, “Did anyone actually test this?”. From hallucinated features to mysteriously “passing” test cases, 99% Confidence, 100% Wrong; celebrates the people who live in the edge cases, break the happy path, and remind everyone that “it worked in the demo” is not a quality strategy.

For every AI that says, “I’m sure,” there’s a QA engineer opening a defect ticket.

Owned Intelligence Is an Estate

Governing a portfolio of models and the agents acting across them

The economics of AI foundation models are quietly redrawing enterprise strategy. As token costs climb faster than returns in high-volume domains, more organizations are choosing to own and operate their own domain models instead of renting intelligence by the call. That decision rarely stops at one model. It becomes a portfolio: several domain models across multiple cost centers, with an agentic layer composing them and acting on what they produce.

Most organizations are earlier on this path than the headlines suggest. The tooling is young, agent development kits are arriving while settled best practices are not, and the journey from experiment to operation is where almost everyone is still working. MIT's research found that over 80 percent of organizations have piloted generative AI tools, yet of the enterprise-grade systems evaluated, only about 5 percent reached production, and the ones that stall tend to stall for the same reason: the systems do not retain feedback, adapt to context, or improve over time. At the same time the surface area keeps expanding. Gartner expects task-specific AI agents in 40 percent of enterprise applications by the end of 2026, while also projecting that more than 40 percent of agentic AI projects will be abandoned by 2027 over cost, unclear value, and weak controls. Whether an organization is running its first pilot or operating a growing fleet in production, the same questions are arriving.

This session reframes that reality around one idea: owned intelligence is an estate. You can buy the models. You cannot buy the behavior of the system they become. An estate has to be operated, and operating it well comes down to two kinds of governance that only work together.

The first is governing the technology. Keeping a portfolio of AI models and the agents on top of them accurate as data shifts, catching drift and hallucination before they propagate into action, and building the reinforcement and human-in-the-loop pipelines that keep domain models honest over time.

The second is governing the organization's use of it. Deciding who owns a model's behavior, measuring whether the estate is producing value, and treating a growing set of models and agents as a workforce that needs onboarding, oversight, and accountability.

Grounded in real deployment experience, this talk gives technology and delivery leaders a practical lens for the estate they are starting to build, at whatever stage they are building it, and the operating discipline that turns a scattered portfolio into a compounding advantage.

What Attendees Will Leave Able to Do:

• Recognize the economic shift moving enterprises from rented intelligence toward owned domain models.
• Separate the two governance jobs owned intelligence demands, and see why the agentic layer makes them inseparable.
• Apply a clear lens for operating a growing portfolio of models and agents, wherever they are in the journey.