Software engineer Brian Przezdziecki uses Codex to turn customer complaints into software fixes, build faster and keep Goliath Data’s real-estate platform improving around the clock from Chattanooga, Tennessee.
Brian first saw the problem at home. His mother was a real-estate agent, and he spent hours in high school helping her with repetitive work. “I wasted so much time just doing the most menial things,” he says.
Real-estate agents and investors still piece together clues about potential home sales from scattered public records, aging databases, spreadsheets and between 10 and 15 apps. Potential home sellers slip through the cracks, since they are not always ready to sell right away. Finding a lead, reaching the owner and staying in touch can consume hours before anyone knows whether a sale is possible.
Goliath Data pulls those steps into one platform. Its software scans tens of thousands of county records each day, connects events such as inherited homes or unpaid property taxes to specific addresses, finds contact information and helps agents identify potential sellers while the information is fresh. AI can contact people who have opted in, qualify their interest, maintain follow-up and schedule appointments.
Goliath applies that same urgency to its own software. Instead of asking salespeople to document customer complaints, wait for engineers and revisit the issue later, the company gives them Codex. A salesperson can describe a broken button, confusing form or other problem; Codex proposes the code change, and an engineer reviews it. The team went from fixing several issues a week to hundreds.
Goliath’s head of brokerage partnerships, Zach Fitch, saw the commercial stakes firsthand when one of his clients, a large single-family investment team, needed to search properties by public-school attendance zone, a filter Goliath did not offer. A salesperson described the request to Codex in a paragraph; over roughly a day, it found public school-boundary data, connected it to Goliath’s property search, built and tested the filter, and enabled it only for that prospect’s account. Six days later, the salesperson demonstrated it in the prospect’s own target area. The prospect asked Goliath to send a $10,000-per-month contract, or $120,000 annualized. Goliath estimates the feature took about $40 in compute to build.
Brian has also configured Codex to work on a schedule, checking company systems, spotting errors and drafting fixes or routing urgent issues to the right person. “Every night when we go to sleep,” he says, the team can give Codex a large assignment and “wake up to a plethora of work to review and merge.”
One July night, a scheduled Codex agent caught a spike in Goliath’s owner-lookup failures around 3 a.m., about half an hour after the dashboard began turning red. As the dashboard’s reported failure rate climbed toward 99%, it worked through tens of thousands of log lines and found two unrelated problems being counted together: roughly one-third came from a background system that had stopped revalidating workflows, while most of the rest were requests from customers with no remaining credits mislabeled as system failures. The agent corrected the workflow, fixed the reporting logic so the dashboard would separate the two cases and documented the incident. By 5:40 a.m., reported failures were back to normal; Brian’s team woke to the incident report rather than an open outage.
Goliath says AI helped it match, in one year, capabilities established competitors spent a decade building. Following a shift to a subscription-and-service model, Goliath estimates revenue grew about 100% month over month for four to five months. The team expanded from three people to roughly 20 full-time employees.
The company is hiring to keep up with its growth, which is driven by AI. One engineer first caught the company’s attention by using AI to build a useful demo over a weekend. He joined as an intern, moved from Los Angeles to Chattanooga and became a full-time employee.
