Agentic Engineering

AI finds the flaws. Can your team fix them in time?

AI models now find software flaws faster than many teams can fix them. Bank of America answered with 1,000 extra contractors. Most companies cannot hire 1,000 contractors. The short answer for your team: know what you run, then fix one important journey end to end, with AI drafting fixes under independent review. Here is what the story means for your apps and APIs, and how to take that first step.

What happened at Bank of America

In October 2026, CEO Brian Moynihan told CNBC (opens in a new tab) that the bank put on 1,000 contractors in the second quarter, on top of 3,000 internal staff. Their job, CNBC reports: patch flaws in the bank's own and its vendors' software, exposed by Anthropic's Mythos model.

He said software fixes and anti-money-laundering compliance together cost the bank a couple of hundred million dollars or more in the third quarter. CNBC does not say how much of that went on patching alone.

Why this is not only a bank story

In May, S&P Global reported (opens in a new tab) that Mythos was forcing banks across the US and Europe to tighten their defenses. Many run fragmented, sometimes decades-old IT systems. CNBC described the problem plainly (opens in a new tab): AI speeds up how fast flaws are found, but companies still take days or weeks to patch them.

At first, access to Mythos was tight. JPMorgan Chase was the only bank among the launch partners Anthropic named. Banks without access, such as ING, worked through vendors that had it. Since October, security teams can apply for access (opens in a new tab) through a vetted program.

But you do not need access to be affected. Bank of America also patched vendor software. The same vendor libraries and shared components may sit in your apps too.

The real gap: finding is fast, fixing is slow

In plain English. A vulnerability is a flaw an attacker could use. Fixing it means changing the code, testing the fix and the features around it, and shipping the fix to every user.

Finding flaws is now fast. Fixing them still takes people who know the code, tests that check the fix, and a release path. That is where teams stall, and usually for good reasons:

  • Legacy systems that few people still understand.
  • Old code with no clear owner.
  • Vendor components you cannot change yourself.
  • Change freezes, audits and sign-offs that exist for a reason.

None of this is a lack of will. It is the real cost of software that grew over years.

Waiting does not shrink the risk

If your software has flaws, they are there today, found or not. AI makes them faster to find, for your defenders and for attackers. Waiting does not reduce your exposure. Every week a known flaw stays open is a week someone else can find it.

The answer is not a rush. It is a small step you can review.

You do not need 1,000 contractors. You need a reviewable first step.

  1. Know your exposure. List your apps, APIs and third-party components. Scan them. Rank findings by how reachable each flaw is and by blast radius: what breaks, and for whom, if this flaw is used. Give each finding an owner. For a single app, our production-readiness audit gives you a prioritized risk report, including security and open-source packages, and a roadmap to fix it.
  2. Fix one journey end to end. Pick the path that matters most, such as login, payments or data export, and fix its most important findings first. Rate each fix by its blast radius, not its size: a one-line change to login can lock out every customer. Any fix with a large blast radius gets an engineer's review, however small. Anthropic's Claude Code team does the same (opens in a new tab): a check flags changes that need a person even when their risk tier says no. Each fix ships with tests and a way to roll back.
  3. Let AI help fix, not only find. AI drafts the patch and the tests. Independent AI review gates check the work, and engineers review where the risk calls for it. AI as a force multiplier for your whole team, not a replacement.
  4. Turn patching into a pipeline. Automatic dependency updates, security checks in every build, and release gates. The next finding becomes a ticket, not a project.
  5. Keep the know-how in house. Specifications, review rules and tests stay with your team, so the next round needs fewer outside hands.

Where we fit

We work best alongside internal teams, from early product concept and architecture to implementation support and launch readiness. For this problem that means: we help you see your exposure, fix the first journey with your engineers, and leave behind the pipeline and the review rules. Since 2001, we have helped teams build mobile apps and product systems, and the review gates in step 3 are how we ship our own code. See how we work with AI agents.

Start a Project Book a production-readiness audit