Each Friday at 5 p.m., Diego Oppenheimer will get an electronic mail that remembers his week higher than he does. It pulls from his calendar, assembly transcripts, and inbox to determine what actually mattered: selections made, guarantees to maintain, and priorities for the week forward.
“It offers me a superpower,” mentioned Oppenheimer, a machine-learning entrepreneur finest often called the co-founder of Algorithmia, who’s now working with startups as an investor in Seattle.
What’s notable is that Oppenheimer didn’t purchase this instrument off the shelf — he constructed it. What began as a private experiment was a problem: might he nonetheless code after years away from writing manufacturing software program?
With the rise of AI-powered coding assistants, he realized he might choose up the place he left off. His private mission, with the unglamorous identify “Actionary,” has grown to someplace round 40,000 traces of what he jokingly calls vibe-coded “spaghetti.” It’s messy however practical.
Oppenheimer’s do-it-yourself AI assistant is greater than a novelty. It’s a window right into a broader shift. People and firms are beginning to hand off items of judgment and workflow to autonomous programs — software program that analyzes knowledge, makes suggestions, and acts independently.
Exploring the agentic frontier
This rising frontier is the topic of Brokers of Transformation, a brand new GeekWire editorial collection exploring the folks, firms, and concepts behind the rise of AI brokers. A associated occasion is deliberate for Seattle in early 2026. This unbiased mission is underwritten by Accenture.
For this primary installment, we spoke with startup founders and DIY builders working to duplicate totally different features of the work of nice govt assistants — coordinating calendars, managing journey, and anticipating wants — to see how shut AI brokers are attending to the human customary.
The consensus: right this moment’s brokers excel at slim, well-defined duties — however wrestle with broader human judgment. Makes an attempt to create all-purpose digital assistants usually run up in opposition to the bounds of present AI fashions.
“I might need my journey agent and my finance agent and my inventory buying and selling agent and my private well being coach agent and my dwelling chef agent, and so on.,” mentioned TA McCann, a Seattle-based serial entrepreneur and managing director at Pioneer Sq. Labs, on a recent GeekWire Podcast episode.
McCann foresees these slim brokers dealing with discrete duties, doubtlessly coordinated by higher-level AI appearing like a private chief working officer.
However even the time period “AI agent” is up for debate. Oppenheimer defines a real agent as one with each autonomy and unbiased decision-making. By that customary, his system doesn’t fairly qualify. It’s extra a community of fashions finishing duties on command than a self-directed entity.
“In case you requested a advertising division, they’d say, completely, that is absolutely agentic,” he mentioned. “But when I keep on with my AI nerd cred, is there autonomous decision-making? Not likely.”
It’s a part of a a lot bigger development. The marketplace for AI office assistants is projected to develop from $3.3 billion this 12 months to greater than $21 billion by 2030. in line with MarketsandMarkets. Progress is being pushed each by enterprise giants similar to Microsoft and Salesforce embedding brokers into office software program, and by startups constructing specialised brokers.
A report by the newsletter “CCing My EA,” citing an ASAP survey, notes that 26% of EAs now use AI instruments. Some concern job loss attributable to AI, however most high EAs see AI as an augmentation instrument that frees time for strategic work.
From summaries to scheduling
One firm exploring this rising frontier is Read AI, a Seattle-based startup recognized for its cross-platform AI assembly summarization and evaluation applied sciences, which has raised greater than $80 million in funding.
Co-founder and CEO David Shim revealed that Learn AI has been internally growing and piloting an AI govt assistant referred to as “Ada” for duties together with scheduling conferences and responding to emails.
Ada replies so rapidly that Learn AI has been engaged on constructing in a delay into the e-mail response time in order that it appears extra pure to the recipients.
Shim has been personally testing the bounds of the expertise — giving Ada entry to a spread of office knowledge (from Outlook, Groups, Slack, JIRA, and different cloud companies) and letting the assistant autonomously reply questions on Learn AI’s enterprise that are available from the corporate’s traders in response to his periodic updates.
“It solutions questions that I might not have the reply to proper off the bat, as a result of it’s not simply pulling from my knowledge set, however it’s pulling in from my workforce’s knowledge set,” Shim mentioned throughout a fireplace chat with GeekWire co-founder John Prepare dinner at a current Accenture reception.
Shim laughed, “I’m keen to take that threat. We’re doing effectively, so I don’t thoughts giving out the information.”
Nonetheless, there are limitations. Ada can wrestle with complicated multi-person scheduling or duties requiring knowledge it may’t entry, and may nonetheless sometimes hallucinate. To handle this, ReadAI incorporates human oversight mechanisms like “sidebars” the place Ada asks for affirmation earlier than sending replies to messages deemed extra delicate or tough.
Shim argues in opposition to the thought of constructing a single, all-encompassing agent.
“The method of brokers doing all the things will not be the correct method,” he mentioned. “In case you attempt to do all the things, you’re not going to do something effectively.”
As an alternative, he believes profitable AI assistants will give attention to fixing very particular issues, very like Google Maps offers driving instructions with out attempting to be a basic journey agent.
The “book-me-a-hotel” problem
Journey is a use case that’s near the guts of Brad Gerstner, founder and CEO of Altimeter Capital. Gerstner is thought for backing a few of the largest names in tech — from Snowflake to Expedia — and for distilling massive tech shifts into easy assessments, similar to his resort reserving problem.
The precise instance he gave on the 2024 Madrona IA Summit in Seattle was telling an AI agent to ebook the Mercer Resort in New York on a particular day on the lowest worth — a typical problem for enterprise vacationers.
“Till we are able to do this, we now have not constructed a private assistant,” he mentioned.
That’s a part of the bigger drawback Michael Gulmann, a former Expedia product govt, got down to resolve with the startup Otto, which is developing an AI agent particularly for enterprise vacationers.
As proven publicly for the primary time at this 12 months’s Madrona convention, Otto tackled Gerstner’s particular problem. After receiving the request to ebook the Mercer Resort on a particular day, it discovered the most affordable out there room, confirmed the worth and particulars, and accomplished the reserving, with minimal prompting, inside about two minutes.
“Who would have thought that Brad Gerstner wished the most affordable room?” Gullman joked.
Otto handles varied features of journey. It understands and learns detailed person preferences — from particular facilities like rooftop bars to most well-liked airline seats, resort room varieties, and loyalty applications — utilizing this information to refine searches and make customized suggestions.
As Gulmann defined in an interview, Otto doesn’t use a single monolithic mannequin. It coordinates a bunch of slim brokers: one to interpret messages, one other to handle loyalty applications, one other to deal with funds. Collectively they simulate a small operations workforce working behind the scenes.
Otto confirms particulars with the person earlier than finishing purchases, despite the fact that it might do this autonomously. Gulmann described that precaution as psychological, not technical — understanding that most individuals aren’t but comfy with AI shopping for issues with out their involvement.
After studying about Otto’s capabilities, Gerstner was impressed and wished to see the way it performs because it strikes into public beta, mentioned Mike Fridgen, a enterprise accomplice at Madrona, which incubated the corporate.
The grand problem of scheduling
If resort reserving is the acid check for autonomous assistants, scheduling conferences is the on a regular basis nightmare.
That’s the issue Howie is attempting to resolve. The Seattle startup’s AI assistant lives within the electronic mail inbox. CC Howie on a thread, and it proposes instances, confirms with all events, creates invitations, and provides assembly hyperlinks.
Howie works from an in depth “preferences doc,” impressed by how skilled executives prepare their human EAs — which cafés are acceptable for conferences, how late is simply too late on Fridays, and so on.
The corporate just lately launched publicly with $6 million in funding and a rising variety of paying prospects. It makes use of a hybrid mannequin: AI supported by human reviewers. That helps keep away from the tiny errors that destroy belief — mixing up time zones, dropping a reputation from a thread, or misreading social cues.
The system simulates selections internally, flags potential errors for evaluate, and escalates something ambiguous to a human earlier than hitting ship.
“If you concentrate on the issues that an important human EA does, software program will not be changing that anytime quickly,” mentioned Otto co-founder Austin Petersmith.
In reality, Petersmith mentioned, a lot of Howie’s customers are human EAs themselves, utilizing it to dump logistics. “No person needs to do scheduling,” he mentioned. “Everyone needs the machines to take this specific job on.”
As fashions enhance, Petersmith hopes Howie can increase into different “meta-work” — the executive overhead that retains data staff from the higher-value actions which are nonetheless the realm of people.
Extra time within the day
For Diego Oppenheimer, this isn’t a hypothetical problem. “I’m extraordinarily calendar dyslexic,” he defined. “I’ll triple-book myself. I’ll comply with go to locations I shouldn’t be. I’ll journey to the mistaken metropolis. Actually dangerous.”
Through the years, he relied on human EAs and a chief of workers to maintain him on monitor. However when he stepped again from operating an organization full-time, hiring somebody simply to handle his complicated, multi-role calendar now not made sense. So he constructed Actionary to assist. It sends the Friday recap to catch him up on the week, flagging points proper earlier than his weekend “reboot.”
Oppenheimer’s mission won the People’s Choice Award at an AI Tinkerers occasion in New York final month. However he’s very clear: Actionary is a private mission, not a product within the making. He developed it for himself, and may’t think about taking over the headache of function requests and technical assist from others.
He’s bullish on the bigger development, and a person and investor in instruments like Howie. However he additionally acknowledges that AI brokers can’t match the excellent abilities and judgment of a human EA, not to mention a chief of workers in a higher-level strategic function.
Oppenheimer’s final aim is extra easy, however nonetheless bold. “I’m attempting to make time within the day,” he mentioned. “That’s what I’m attempting to do.”
GeekWire’s Todd Bishop reported and wrote this text with modifying help from AI instruments together with Gemini and a customized OpenAI GPT educated in GeekWire’s editorial method. All info, quotes, and conclusions had been reviewed and verified previous to publication.
