Praxent knows what your bank forgot

A payments system can be obsolete and indispensable at the same time. In one Praxent engagement, an unnamed global processor depended on a Delphi application reconciling more than $2 billion daily. The original engineers had left. Documentation had vanished. The software kept doing its job. Explaining that job had become the problem.

That predicament explains the appeal of Praxent, the Austin consultancy founded by Tim Hamilton in 2000. It works where a better interface must coexist with underwriting rules, settlement routines, and people who need to approve a change. Much of the value is below the screen.

The machine still worked.
The memory had gone.

In the payments case, support had eroded around a functioning application: expired compiler licensing, an unsupported operating system, and hardware that could not readily be replaced. Praxent used its Decode agent system to recover requirements from the code and send unresolved questions to human experts. Software archaeology acquired a faster shovel.

The cost comparison deserves careful reading. Praxent puts traditional discovery at roughly $250,000 in engineering time; its AI approach used approximately $2,500 in tokens. Tokens are one input. The comparison does not establish a complete project price, including engineers, validation, and migration.

The distinction is the story. Cheap computation changes which investigations are worth attempting. Humans still have to decide what the recovered rules mean. A faster answer is useful only if someone can check it.

A small slice earns a bigger yes

At banking technology provider DCI, the difficulty was trust. GoBanking contained years of embedded behavior. Earlier design handoffs required engineers to reconstruct static screens, while unstructured AI prompting lacked consistency. Praxent introduced shared specifications, business-rule identifiers, test gates, and human review.

In its account of the engagement, Praxent reports two end-to-end features delivered in eight weeks, against an initial plan for one. DCI also received a 41-component design system. These are project results, rather than a promise that every banking rebuild will move at the same speed.

“We believe technology should support human expertise, not replace it.”Daren Fankhauser · DCI

The practical change was a process DCI’s engineers could adopt. A narrow feature made the method inspectable. Shared specifications reduced translation between teams. Review gave the client a reason to trust the output.

NEWITY offers the commercial version of the same idea. Praxent’s work began with a borrower portal and loan servicing, then expanded into a broader small-business lending platform. Its engineers built an orchestration layer connecting more than 16 fintech APIs, giving separate services a common customer experience.

Praxent reports a 13-week launch for NEWITY’s pre-qualification experience and 27 underwriting hours saved per week. Those measures concern different things: delivery speed and operating effort. Together they show what a lender actually buys when it commissions software: less friction between an application and a useful decision.

The people behind the permissions

Hamilton started the company as a student. Today its services extend from UX research and custom web and mobile development to data transformation, AI agents, and workforce enablement. The customer list includes Origence, Ascensus, Ameritas, and Dimensional Fund Advisors. These are institutions buying engineering judgment alongside implementation.

Its published culture supplies an interesting constraint on that promise: a workweek of no more than 40 hours. Careers materials describe career ladders, sponsored certifications, and a peer-nominated celebration called the Praxies. The premise is that repeatable delivery should come from a working system, rather than permanent overtime.

The AI offering has a similar preference for defined responsibilities. PraxentGuard is its controls framework, advertised with 29 security checks. Claude Cowork enablement turns procedures into reusable skills; agent engineering adds permissions, monitoring, and review. These are services for organizations that must operate the result after the demonstration ends.

Partnerships help supply the machinery. Praxent identifies itself as an Anthropic Select Tier partner and a Microsoft Cloud and AI solutions partner. Its WealthBlock relationship addresses another practical seam: customized investor onboarding and capital-raising experiences built around an existing platform.

Buy a result. Keep the method.

There is a useful ownership question here, too. An institution can commission a working feature and still remain dependent on the team that built it. Praxent’s stated handoff includes coaching alongside delivery. Buyers can make that concrete by asking who will maintain the specifications, approve future changes, and investigate a failed test once the outside engineers leave.

Praxent sells business-to-business engagements. Its current outcome-based offerings start with an agreed metric, such as funding speed or servicing effort. The company says clients own commissioned work, with a permanent license to relevant accelerators. Clutch lists a $25,000-plus project minimum and a $50-$99 hourly band. A specific engagement still needs its own scope.

In April 2026, Delta-v Capital made a strategic growth investment to expand the firm’s AI capabilities and team. A September announcement added Ranjit Tinaikar, Jeremy Wing, and Garrett Marsilio to its board. The direction is clear: more capacity for financial institutions moving beyond experiments.

For a buyer, the alternatives include internal engineers, broader consultancies, and packaged platforms. Praxent fits when existing financial behavior and a bespoke customer experience must be reconciled. A standard product deserves consideration when it already meets the need.

A sensible first meeting would bring a workflow, its current bottleneck, and a baseline measure. How long does an application wait? Which questions generate support tickets? Where does a person copy the same information twice? Those questions give the engineering brief a destination. They also make it possible to judge a proposal without treating the number of AI features as evidence of progress.

An agent also needs a boundary around its authority. For a financial institution, faster document handling and an automated credit decision carry different consequences. Praxent’s approach puts human approval, identity controls, and action logs into the architecture. The buyer’s job is to name the decisions that require judgment before handing over the repetitive work. That discussion belongs in the brief, alongside delivery dates and budget.

The transferable discipline is pleasantly unglamorous. Pick a feature small enough to verify. Recover its rules. Ask experts about exceptions. Test before expanding. This depends on access to code, people, and evidence; without those, speed compounds uncertainty. The best opening question for a rebuild may be: what does this system know that we have forgotten?