Beyond the chatbot: At the GeekWire conference, AI leaders say the era of autonomous agents has arrived.

The debate over whether AI will transform industries is over.

At GeekWire’s Agents of Transformation conference in Seattle on Tuesday, founders, executives and engineers in attendance tackled the tough questions — what’s working, what’s not, and how fast everything is moving. The thread running through almost every conversation was the move from AI as a conversational tool to AI as an autonomous actor – software that doesn’t just answer questions but acts on its own, evolving as it goes.

Speakers from Microsoft, Amazon Web Services, OpenAI and elsewhere described a world where the barriers that have defined their work for decades are being eliminated, and where the biggest obstacle to taking that value is not technology – we are finding a way to redesign work processes and organizations that were not built with any of this in mind.

Agents of Transformation is presented by Accenture, and builds on GeekWire’s ongoing editorial series, also underwritten by Accenture, highlighting how startups, developers and tech giants are using intelligent agents to innovate.

Read on for quick summaries and key takeaways — with the help of AI, of course — from each fireside chat and panel discussion.

Charles LamannaMicrosoft’s senior vice president of Business Applications & Agents, revealed a moment that caught everyone’s attention: an AI agent turned down 17 meetings on his behalf. He didn’t summarize them, he didn’t flag them – he refused.

For Lamana, that was the moment when AI crossed over from information acquisition into real action. The era of AI as a conversational assistant, he bluntly argued, is behind us. “The sun has set.”

Three important points:

• Don’t invent new AI metrics. The biggest trap Lamanna sees companies falling into is building elaborate AI systems that are disconnected from business outcomes. His rule: use the metrics you already have — revenue, retention, customer satisfaction, supply costs. “No human business metrics should be invested in 15 agents,” he said. If AI doesn’t move a number the CEO already cares about, it’s a hobby.
• Give everyone a good AI, focus on a few big bets. Successful AI transformations share two characteristics: broad access to tools for all employees, and a handful of high-priority projects that are tracked from top to bottom. Companies with 250 “Gen AI” projects are a red flag, not a success story.
• The token budget is a new equation. Lamanna’s teams are already measuring AI spend per engineer as a hiring factor — and candidates are negotiating it. Another engineer told him that he would only take the job if his team had enough tokens for the day. “If you hire a developer who’s lived through this code-agent approach and you tell them your token budget is $1 per day,” he said, “they’re going to be like, ‘see.’” (Read more on that point here.)

Julia Whitethe CMO of AWS, has spent nearly three decades in marketing — and says his biggest challenge right now is not learning too much of it. The principles he proposed years ago, such as one-to-one marketing on average, are now back on the table.

“Every day I have to stop and let go of the things I thought were true,” he told the president Andy Taythe global lead for Accenture Cloud. The obstacles that made those dreams impossible are gone.

Three important points:

• Let it rip – by choice. White’s team sends out thousands of emails a month, and for years each one needs to be signed by someone before it goes out. Since then they have developed a supervised process that has gradually gained the trust to eliminate that step altogether. Meanwhile, experiments using AI for highly productive TV ads taught them the same thing about failure – they took what worked and applied it to digital display ads, going from about 100 variations to many more, almost effortlessly.
• Start with what people hate. The fastest way to buy a group isn’t a big conversion project – it’s getting rid of the little annoying things. White demoed a new all-hands content workflow that cuts a three-hour publishing process to 30 minutes. The room erupted in spontaneous applause. “That’s a pretty high bar” for technology rollouts, Tay added.
• Hire people who don’t know the rules. White said he is deliberately hiring more new students than ever before – people who have no idea how sales has always worked. His idea: New eyes don’t need to learn anything.

Deepak Singh he’s spent nearly 20 years at Amazon Web Services building tools for software developers, and his four-word summary of his daily routine says everything about where things stand: “I live with agents.”

The VP behind Kiro, Amazon’s AI-powered developer site, uses four agents every day — one for research, one for writing in his personal style, one for email processing, and one for internal documentation. It’s not a demo. The way he actually works.

Three important points:

• How you receive is more important than whether you receive. Amazon’s internal research of 40-50 engineering teams found a big difference: teams that tied AI agents into existing workflows gained 20-40% faster. Teams that reorganized their entire environment around agents – cleaner code-switching packages, better documentation, clearer instructions – got 3 to 10 times faster. The difference was not the tools. It was a setup.
• Your monitors are designed for people. Singh’s sharp point about agent security: every policy and permission in your organization is designed for human speed. Agents don’t get tired, they don’t give up, and they don’t stop to call for help – they keep going, which means they can repeat the same mistake a hundred times before anyone notices. Human-designed permutations need to be completely rethought for non-sleeping systems.
• Use them at home, not just at work. Singh’s closing advice went further than most: don’t just outsource agents professionally, live with them personally. The more fluent you are, the more you will stand out when it matters.

Three doctors who spend their days in the messy field of deploying AI — not selling it, actually doing it — keep coming back to the same uncomfortable theme: the technology is the easy part.

Angela Garinger access, Jeremy Tryba of AI research nonprofit Ai2, and Liat Ben-Zur For LBZ’s advice each watched promising AI releases not because the tools failed, but because the people around them did. The panel was moderated by Emily Parkhurst of Formidable Media.

Key highlights:

• Little bits are wide, all the time. The panel agreed that companies that herald a sweeping AI revolution across organizations are the most likely to fail. Winners are operated on – choosing one boring job, applying an agent, estimating the result, and measuring it. “Those who are really successful are very conscious of which high-friction workflows they want to take first,” Garinger said.
• Fear is a real barrier to adoption. Ben-Zur described a pattern he sees all the time: the pilot works well, the monitors like it, and then the rollout just … stops. When teams dig, the reason is almost always fear – fear of being replaced, fear of being judged when a tool goes wrong.
• Clarity unlocks everything. Tryba described watching even top researchers hesitate to use AI tools because they weren’t sure what they were allowed to do with them. The fix was simple: a clear matrix of approved uses, posted to Slack. The next day, everyone had signed up. Consent, it turns out, is a function of coercion.
• Track meaningful metrics. Leaders love to tout the hours saved and what percentage of employees use AI, but Ben-Zur said they need to look at the metrics they’ve always valued — improved revenue, higher retention, better performing features. “I wouldn’t measure how many hours people save – like, ‘Joey saved five hours.’ I don’t care. What does that mean for business?”

Vijay Rajithe CTO of applications and head of engineering at OpenAI’s new office in Bellevue, has a signature move: opening his laptop in meetings so Codex — the company’s AI coding tool — can continue building while he’s away from his desk. It’s an apt metaphor for how you think about AI right now – always working, always integrating. Meta veteran and founder of A/B testing company Statsig talked about what living on the edge actually looks like.

Three important points:

• Everyone is a builder now. Raji built his own personal Slack and email shortcut – local, cloudless, no security – in an afternoon using Codex. His point: the barrier to creating custom software is broken. “Everyone will be a builder,” he said.
• Overcapacity is a real problem. The models are advanced in the way that most people use them. Raji calls this the “skill gap” – and people who bridge that gap, he said, are already more productive than those who haven’t realized it’s there.
• Engineers become agent managers. The next wave isn’t just AI-assisted coding — it’s a bottleneck. The productivity gains from AI are now so fast that the new limit is for humans to review all incoming code. The title of the future job, he suggested, is actually “manager of agents.”

Thanks to presenting sponsor Accenture; gold sponsors Nebius and AWS Marketplace; and silver sponsors Core Team Partners, Astound Business Solutions, OneByZero, Autessa, Pay-i, GemaTEG, Cascade, and WTIA for helping make the event possible.