A travel app has five people on its engineering team and four AI tools in circulation. You might expect a small explosion of productivity. At Travly, according to Arbisoft’s account, individual engineers got faster while the team remained stuck. The missing ingredient was surprisingly ordinary: nobody had written down enough of what everybody knew.
- Arbisoft builds software and embeds teams inside customer businesses.
- Education and travel supply some of its deepest experience, including edX and KAYAK.
- Its practical lesson: preserve product knowledge before asking tools to accelerate it.
Database routing rules, sensitive pricing branches and supplier quirks lived in people’s heads. Each assistant arrived without that memory. The code was available; the reasons behind the code were scattered. This is a useful place to begin understanding Arbisoft, a company whose work often sits behind a product name you already recognize.
A partner measured in years
Founded in 2007 by Yasser Bashir and Umair Shah, Arbisoft sells the people and expertise needed to build, repair and operate software. Its engineering base is in Lahore, with a listed headquarters in Plano and other international offices. Customers can add specialists to their own teams, commission a dedicated squad or bring in help for a defined project.
That places it among software services firms, where alternatives include hiring internally or contracting another development company. The interesting distinction is the length and depth of particular relationships. Arbisoft’s edX work began in 2013. In 2016, Open edX itself described a team of more than 50 Arbisoft people working across development, customization, course authoring, testing and support.
A learning platform contains a remarkable amount of plumbing. Someone must make course imports behave, connect enterprise sign-on, handle assessments and help a learner when something breaks. Arbisoft’s edX case describes work across those layers, including Studio authoring features, analytics and integrations with external learning systems. The visible course is only the front room. The rest is a network of dependencies that has to work for administrators, instructors and learners at once. A change that pleases one group can create extra work for another.
KAYAK supplies another example. Arbisoft describes an engagement involving mobile development, data extraction, quality assurance and automation, organized into specialized teams. A flight-search app must survive much more than a successful demonstration. Prices change, suppliers behave differently and devices disagree. Knowing yesterday’s awkward exception can be useful when tomorrow’s bug arrives.
The argument for an embedded partner follows from that accumulation. Replacing people has a cost beyond recruitment: somebody must learn the product again. A buyer assessing Arbisoft should ask which experienced people will actually join the engagement, how they communicate and how their knowledge survives a handover. A famous customer logo cannot answer those questions.
The bill behind the button
Arbisoft’s work also reaches business decisions. In its edX mobile payments case, learners could browse courses in an app but had to leave it to buy an upgrade. That detour created friction. Rather than immediately building everything, the team tried a painted-door test: an upgrade button that measured interest, followed by an A/B experiment.
Then came the arithmetic. The case describes an initial 30 percent app-store transaction fee and the use of reduction programs that brought the relevant fee to 15 percent. Those are historical project details, not a universal statement about today’s store rules. They explain why a better purchase experience still needed a viable commercial model.
The team began with 50 courses. Arbisoft dates the expansion to all courses to February 1, 2024. This is the part another product team can copy: test demand, examine the margin, release a manageable slice and use what happens to decide the next step. Interest in a button is evidence; it is not yet a business.
Arbisoft’s own services are sold through proposals. Its inquiry form groups prospective budgets from below $50,000 to above $500,000. Those boxes are a way to qualify projects, not a rate card. Buyers are paying for scope, staffing and delivery responsibility. Its Edly learning-platform business adds a different model, with pricing tied to monthly active users.
A week, compressed
Sometimes the product problem is easier to see with a clock. Advanced Energy United needed fresh public-utility information for InsightEngine. Arbisoft says the existing collection process took 168 hours. A service needing daily updates was waiting roughly a week for its raw material. The bottleneck arrived before anyone opened a dashboard.
At least 97.6% less waiting. Same underlying job: get useful data to subscribers.
The response combined distributed crawling, intelligent extraction and cleaning data as it arrived. An API fed the customer-facing application. Arbisoft reports reducing runtime to under four hours. This is engineering as a change in operating rhythm: information becomes useful sooner. The case illustrates its data expertise more clearly than a long list of programming languages could.
The handbook before the agent
Return to Travly. Arbisoft documented conventions in repository files, assigned agents distinct roles with human owners and built reusable instructions for recurring work. It also changed specification writing, including a critic stage to catch vague requests. The system went live on March 17, 2026; the company reports roughly three times the sprint output over the following 60 days.
“If a process increases trust, choose the process. Otherwise, choose trust.”Arbisoft’s published values
That result is a company-reported case, not a promise for every team. The setup also required weeks of preparation. If nobody can explain pricing exceptions, review a specification or take responsibility for an agent’s changes, automation has poor instructions to follow. The scarce resource shifts toward judgment, and judgment still needs an owner.
Arbisoft has reasons to care about that ownership. Alongside client delivery, its venture studio builds and operates products, including Edly. Open-source foundations offer customers control, while customization, upgrades and support remain real work. The useful buying question is concrete: who will understand this system when it changes? Arbisoft’s strongest cases make a persuasive argument for keeping that understanding close.