The mabl blog: Agentic Software Testing

Beyond Friday the 13th: Building trust in the enterprise | mabl

Written by Izzy Azeri | Nov 19, 2026, 4:39:45 PM

The week software made the news

Last week, roughly 250 people in Chicago tried to push the Bean seven inches left to free a man supposedly trapped inside. It didn’t budge, making it the most resilient system in the news.

City officials confirmed nobody was inside the Bean, yet the software outages dominating the same news cycle were real. Microsoft confirmed an Outlook bug dormant since July until a Thursday traffic spike disrupted email into Friday the 13th. That afternoon, the FAA halted departing flights nationwide for over four hours after a routine router upgrade. By Monday, a software bug slowed transactions at JPMorgan Chase, delaying payments elsewhere too.

Each incident was traced to an ordinary software or configuration change. The software factories behind modern delivery, with their coding agents and automated pipelines, push out more change than teams can verify, leaving enterprise software more vulnerable to failures of every kind.

Faros AI’s Q3 report found that at peak AI adoption, monthly incidents rose 125% and unreviewed pull requests jumped 76% over earlier quarters. Every incident pulls top people off strategic roadmap investments, and severe ones erode customer trust.

Why verification has to be independent

Coding agents amplify the risk of future newsworthy headlines because they are built to finish tasks. Columbia’s DAPLab built over 15 applications across five leading coding agents and found they prioritize code that runs over code that is correct, often suppressing errors rather than surfacing them. Ask that same agent to test its own work, and it writes checks agreeing with its implementation without real business signal.
Regulated industries like banking and pharmaceuticals have long separated execution from sign-off through governance and healthy dissent. Software engineering requires that same discipline.

Independent verification requires a dedicated testing harness: the execution framework validating releases separately from the coding agent, with no stake in pushing changes through. Effective verification carries business context: critical customer journeys, quality standards, and historical failure patterns. Architecture matters as much as the model; Stanford and MIT researchers found the same model performs up to six times better or worse depending on the surrounding system.

Skipping independent checks simply shifts the cost. Another data point from Faros AI shows QA time surged 300%, proving skipped review checks get absorbed later manually or by customers in production.

Scaling independence across the enterprise

Notre Dame professor Mike Chapple, quoted in The New York Times, captured the common thread in last week’s outages: the systems we depend on every day are complex and fragile.

Independent verification delivers meaningful impact only when validating real workflows end to end at enterprise scale. We are grateful to partner with organizations tackling this complexity daily. Here's how mabl customers put it into practice:

  • Global Airline: Validates a multi-step loyalty booking for a family of three every night, running 1,300+ steps across 18 scenarios in 15 minutes instead of 2.3 hours sequentially.

  • Property Insurer: Ships 1,300+ deployments monthly, with each release automatically triggering roughly 27,000 browser and API runs without manual intervention.

  • Multi-Brand Home Goods Retailer: Tests storefronts across eight brands from shared data sets, executing 265,000+ runs in 30 days while mabl auto-heals nearly 79,000 UI element changes.

Building that capability in-house means owning browser grids, private-network runners, brittle selectors, and failure triage. Without sustainable maintenance, teams lose trust in their test suite — and the green checkmark loses its strategic value.

Evaluating your software factory for the agentic era

As leaders invest in a software factory or an agentic development lifecycle (ADLC), where autonomous agents assist across planning, coding, and deployment, the core question is governance.

Ask what independent system verifies your output, and whether it scales alongside your agents. Quality isn't just asking, "Does the code look right?"; it’s knowing, "Can a customer still log in, pay a bill, or book a flight?".

The next Friday the 13th arrives in August. Now is the time to invest in the systems and quality culture protecting your customers.