Axos Financial (AX): AI Coding Tools Hit 90% of Committed Code

On its 2026-04-30 earnings call, Axos Financial, Inc. (AX) put a number on the AI coding tools its own engineers use for the first time: artificial intelligence accounted for 90% of committed code [1]. On the 2024-07-30 call the company had said only that Microsoft Copilot was used across all of its development [5]. Across the disclosures in between, the story moved from naming a vendor to reporting an adoption rate.

Axos Financial, Inc. (AX) is a bank. It offers customers banking products such as securities-backed lending and secured credit cards, and it runs a securities custody and clearing business that serves broker-dealers and registered investment advisers. The AI described here is not sold to customers. It is tooling the company's own software engineers use in day-to-day development: Microsoft Copilot first, then further AI-enabled tools management has not named, and by 2026 agentic tooling that can run through a process on its own. The tools generate and complete code, lift and shift legacy code written in old languages such as BASIC into new environments and refactor it, read old code and write its documentation, turn plain-language business requirements into task lists a developer can work from, and review and update code before it is committed. "Committed code" here means the code engineers finish and formally merge into the company's code base. The users are only Axos's own engineering team, which makes this a support function.

How the disclosure evolved

Each period has been harder than the one before it. On the 2024-07-30 call, management said in Q&A that Microsoft Copilot was used across all development and that other unnamed AI tools were also speeding development up. That gave a vendor but no task, and no productivity or headcount figure of any kind [5].

The 2025-04-30 call inverted that. Management broke the workflow into three concrete jobs - migrating and far faster refactoring of legacy code in old languages, automatically extracting and writing code documentation, and turning plain-language requirement documents into development tasks - but named no tool at all, and did not say whether the three were in production or in testing inside the company [4].

On 2025-10-30 the subject moved out of Q&A and into prepared remarks for the first time. Management used the present tense to say AI was affecting the company's efficiency and software development, but improvements in speed, quality and cost were still hung on further AI implementation [3].

The 2026-01-29 call was the most complete qualitative disclosure in the window, and the only time management named what AI was meant to replace: deploying AI tools across the entire software development life cycle to review, document and update code, so that the team could take on more projects concurrently at a faster pace with fewer resources, without having to increase the pace of new hires or offshoring [2].

Earlier calls did not mention AI. That means there was no disclosure at the time; it does not support an inference that the company was not using these tools then.

What the latest quarter added

The most recent period gave numbers for the first time, and what they measure is adoption intensity. The 2026-04-30 call said that since the beginning of calendar 2026 the number of technical uses of AI tools had increased by 37%, raising artificial intelligence's share of committed code to 90%. The same call also wrote agentic tooling into the disclosure for the first time, used for testing, automation and quality checks [1].

Both figures can only be read as how widely the tools are being used. They do not say that AI writes 90% of the company's code. The disclosure does not state whether that share means code AI wrote, code it assisted with, or code it only reviewed before commit, and it does not say which repositories or teams are covered. "Technical uses" is never defined and could be users, use cases or invocations. The 37% comes with no starting base. The measurement window is a partial period management chose itself and is not aligned with the fiscal quarter, so the two numbers cannot corroborate each other.

Further down the chain this lands on non-interest expense, and within it on salaries and benefits and on professional services. An engineering team of the same size running more projects in parallel means incremental development capacity does not have to be bought with new engineering hires, and purchases of offshore and outsourced development would fall. For now that link rests on management's statements alone. The latest period bundled AI with other efficiency measures in a single sentence carrying no numbers, and non-interest expense in that period actually rose sequentially. It can therefore only be written down as a possible relationship, not as a confirmed cost saving.

What to watch next

This history currently supports one judgment: Axos is betting AI on the substitution of engineering labor. The one time management named the resource AI was meant to replace, on 2026-01-29, it was the pace of new hires and offshoring. In the period after that, all the company could produce was still an adoption rate, with no evidence of substitution being realized, and non-interest expense in the latest period rose sequentially instead.

For a reader, AI at AX is a cost assumption that has not yet reached the financial statements. Whether it holds will be written on the salaries and benefits and professional services lines, not in the 90% figure.

What would genuinely flip that judgment is the direction of engineering headcount and of offshore and outsourced development spend over this window. Management itself framed not having to raise the pace of hiring and offshoring as the point of this AI. Only if revenue keeps expanding while those two stop rising alongside it will the adoption rate have connected to cost for the first time.

Application assessment

• AI coding tools used inside Axos's own software development life cycle - Microsoft Copilot first, then further unnamed AI-enabled tools and, by 2026, agentic tooling - applied to the engineering function itself rather than sold to any customer. | Business position: support function | Deployment stage: limited production | Scope: single business unit | Value type: cost reduction

Sources

[1] Drillr · Axos Financial, Inc. (AX) · 2026-04-30 · earnings call

"Since the beginning of calendar 2026, the number of technical uses of artificial intelligence tools has increased by 37%, increasing artificial intelligence's share of committed code to 90%."

[2] Drillr · Axos Financial, Inc. (AX) · 2026-01-29 · earnings call

"These AI-enabled tools allow us not only to review, document, and update code at a faster pace with fewer resources, but they will also allow our team to take on more projects concurrently without the need to increase the pace of new hires or offshoring."

[3] Drillr · Axos Financial, Inc. (AX) · 2025-10-30 · earnings call

"AI is having an impact on our efficiency and software development."

[4] Drillr · Axos Financial, Inc. (AX) · 2025-04-30 · earnings call

"The ability to lift and shift old code that is in basic or in some sort of old language and be able to refactor it much more quickly or to be able to extract and document the code which would normally require somebody who is very skilled at reading the code and documenting it and all those things."

[5] Drillr · Axos Financial, Inc. (AX) · 2024-07-30 · earnings call

"Copilot from Microsoft utilized across all of our development is also another area."