Before Omar Abdelaziz tried to give the iPhone a pair of invisible hands, he used his own. In high school, he bought used iPhones, fixed them and sold them to local customers. The operation began at roughly one device a week in 2012. It grew into repairs for individuals and shops, plus Apple-product resale around the world. His public career record says the business eventually cleared $100,000 in trailing 12-month sales. The company name was gloriously literal: iSellAppleGear.
Fourteen years later, the same object is back at the center of his working life. Blue, the company Abdelaziz co-founded and runs, pairs an iPhone app with a little USB-C device. A person speaks. The assistant opens apps, reads what is on the screen, taps, types and swipes. It is meant to finish the fussy work trapped between services: locate the receipt, upload the expense, answer the message, book the thing.
The symmetry is almost too tidy. First he repaired the phone. Then he spent six years helping Google improve what a phone could understand. Now he is trying to make it act.
A résumé full of friction
Abdelaziz studied finance and computer science at Washington University in St. Louis. Before graduation, his jobs skipped cheerfully across industries: law, accounting, tax, financial analysis, consulting, retail innovation and investment risk. The interesting thread is less the variety than the recurring nuisance he chose to remove.
At a St. Louis valet business, he built a reporting system with Google Forms, Sheets and scripts. His profile credits it with saving management more than 100 hours each month. During a tax internship in Cairo, he wrote a script to automate a payroll-email process and reported saving 3,600 hours a year. He also created Participatr, an app for classroom participation, and recruited campus representatives to sell it to professors.
The finance-and-computer-science pairing suits this record. One discipline asks where value goes; the other offers a way to rearrange the route. Abdelaziz reported a 3.9 GPA in each field and a 3.8 overall average. In 2023, while at Google, he also spent a term as a part-time graduate student in artificial intelligence at Stanford Engineering. The formal study arrived after years of informal practice: find the bottleneck, put a number on it, then see whether software can make the number smaller.
This is an operator’s version of curiosity. The broken process is not beneath consideration; it is the invitation. A spreadsheet deserves attention if it steals an afternoon. A payroll routine is interesting if a script can give people back their week. Glamour is optional. Repetition is the clue.
Inside the assistant factory
Google hired Abdelaziz after university in 2018. His first product role involved last-mile navigation data in Maps. He says he patented an approach for guiding drivers to available street parking. He also recruited and led a 25-person group of employees contributing part time to launch Google Maps Snake, the 2019 April Fools’ game that turned cities into bright little train routes. It remains an appealing footnote: even serious infrastructure people occasionally need a snake.
The next roles moved closer to voice. He worked on Google Assistant for Android in India, then Voice Search in the Android Google app. In his public career account, he describes himself as the product’s number-one user and credits his team with several of the app’s leading launches in 2021. The language is revealing. He did not merely study the user. He tried to become an unusually demanding one.
“I’ve spent more than a half-decade leading teams on Google Assistant and at Scale AI.”Omar Abdelaziz, on his YC founder profile
By 2022, he was pitching a screen-based agent inside Google: give an AI a goal, let it read the display, and have it operate the phone without relying on a custom integration for every app. The project was canceled after a reorganization, according to his profile, but the central idea survived. He then worked on app search and a project to move Google Assistant from traditional language systems toward a large-language-model architecture. That included data generation and automated model evaluation, the quiet machinery required to make a demo behave more like a product.
His voice work also left a public paper trail. In 2020, Abdelaziz and eight collaborators published a defensive disclosure about evaluating automated speech transcription through user feedback. The problem is easy to state and awkward to solve: checking a machine transcript often requires someone else to hear what the user said. Their publication explored feedback signals as another route to measuring accuracy. Years before Blue’s keyboard tried to clean up a spoken thought, he was already working on how a voice system could learn whether it had listened well.
The long route to one small plug
After Google, Abdelaziz spent six months as a staff product manager on Scale AI’s generative-AI machine-learning team. His description of the role again centers on workflow: automated checks for data labelers, automated grading and an agentic system meant to improve quality, cost and speed. The stage changed. The favored move did not.
There was a human-scale detour, too. In 2021 he co-founded Street Mama, a short-lived San Francisco nonprofit that reported distributing more than 4,000 meals. It sits oddly on a conventional technology résumé and comfortably in this one. Coordination, delivery and repeated small tasks are also systems problems, except the spreadsheet has an immediate consequence at the other end.
The plug is the point
Blue joined Y Combinator’s Summer 2025 batch with co-founders Farhan Hossain and Peter Krogh. The team’s backgrounds fit together with unusual neatness. Abdelaziz brought Google Assistant and computer-use work. Krogh had spent years designing spoken dialogue and, most recently, actions for Gemini Live at Google DeepMind. Hossain had worked on hardware and systems, including Apple Vision Pro.
Their product argument is simple enough to repeat at dinner. A voice assistant can be clever and still fail at the last inch. It may know what ought to happen, yet lack a way to press the button. Blue’s small hardware layer gives the system a route through the interface already on the phone. The assistant uses the apps a person has rather than waiting for every app maker to construct a special doorway.
In a public demo, Blue moves from Slack to Google Docs to an expense tool. It reads messages, drafts a reply, changes a document permission after a spoken correction, and uploads a receipt from recent photos. The valuable moment is not any single trick. It is the handoff between them. Modern work is mostly handoffs wearing different icons.
Ask → observe → clarify → approve → done
Abdelaziz has tested the idea in deliberately ordinary situations. He has described configuring Blue to review work-card charges around 2 a.m., find receipts in Gmail, Amazon or Uber, and upload them to an expense app with a note. During an early rollout, he offered to hand-deliver devices to Bay Area real-estate agents. He also replied directly when early users posted problems. The chief executive’s office, in this version, includes delivery driver and support desk.
In early 2026, he said the company had shipped about 200 orders in one week and removed its waitlist. By May, Blue had added a second wedge: a smart dictation keyboard for iPhone. The keyboard does not aim for strict transcription. It sends audio to a model and asks for the text the speaker likely would have produced after reviewing and editing. A false start can disappear. A repeated phrase can resolve. A person can change direction halfway through a sentence without making the recipient read the wreckage.
“People don’t use it three days a week, one time for two minutes. They use it 100 times a day.”Abdelaziz on Blue’s dictation habit
The ambitious part is mundane
The broad promise of an AI that operates a smartphone naturally attracts questions about reliability, approval and trust. Blue’s visible interface offers one answer: the user can watch the work occur and intervene. Abdelaziz has said publicly that the product runs in the cloud and that humans do not view what users do. For a product built around personal apps, those operating details are not footnotes. They are part of the product.
Abdelaziz describes his aim in sweeping terms: helping define the next era of computing, beginning with a much more capable voice assistant. Yet the company’s demonstrations wisely stay close to the ground. Nobody needs a lecture on the civilizational importance of filing a receipt. They need the receipt filed.
That practicality has been present since the repair bench. A cracked phone, an awkward spreadsheet, a maintenance-mode search product, a pile of unlabeled expenses: each is a small affront to time. Abdelaziz’s career keeps returning to the same bargain. Give the system more of the tedious motion so the person can keep the judgment.
The little blue plug makes the idea almost comic in its physical modesty. Years of speech research, model evaluation, interface observation and hardware engineering arrive as something you can misplace between the sofa cushions. But perhaps that is exactly right. The next era of computing rarely announces itself as an era. Sometimes it plugs into the bottom of the phone and starts tapping.