Major national retailerRetail
Legacy ETL converted to Databricks
78%
of legacy transformation types handled by the AI-assisted migration tool
Built onDatabricksRead the case study
AI
How we build with AI
Our engineers build with Claude and other current models. AI agents do the build work, a second agent reviews it adversarially and an engineer signs off before anything ships.
The method
AI builds it. A second agent attacks it. An engineer signs it off
Two of the five steps belong to people, and they sit where judgment matters: before the work starts and before anything ships.
- Engineer01
Scope and baseline
An engineer defines the change and pins down how the system behaves today, with the tests, scanners and dependency maps. Where tests are thin, they are written first.
- AI agents02
Build
Claude and other current models do the build work: upgrades, remediation, conversions and new code, in small changes an engineer can review.
- Review agent03
Attack the change
A separate agent tries to break each change: bugs, security gaps, behavior that drifted from the tests. Results are also checked against an independent source, such as a second vulnerability database.
- Engineer04
Sign off
An engineer reads the change, the test and scan results and the review findings, then approves it or sends it back.
- Your team05
Ship with the evidence
The change deploys with its record: passing tests, scan results and a plain-language summary your engineers and leadership can both read.
Review findings and rejected changes go back to the build agents for another pass.
Why it holds up
AI carries the volume. Engineers own the decisions
AI makes the slow, high-volume work practical: upgrades, remediation and conversions that used to sit in a backlog. It does not make judgment optional, so every change reaches an engineer with the evidence already attached.
No model grades its own work
A separate agent reviews each change, asked only to find what is wrong with it.
The model matched to the task
Claude and other current models, matched to the task, with private or local models available where code cannot leave.
Tests before changes
Current behavior is pinned down first, so every change is checked against what the system already does.
Evidence your team can read
Passing tests, scan results and a plain-language summary at each milestone.
Case studies
The method on client work
How our engineers work on the engagements where we use it: security remediation, framework upgrades and codebase consolidation for a national distributor, and legacy ETL conversion for a major retailer.
AI-assisted modernizationNational distributorDistribution
Customer portals secured, upgraded and consolidated
- 620+
- known vulnerabilities eliminated
- Angular 16 to 21
- and .NET 6 in-process to .NET 10 isolated
- $25K+ a year
- realized, about $78K a year projected from consolidation
Major national retailerRetail
Legacy ETL converted into tested Databricks pipelines
- 78%
- of legacy transformation types handled by the AI-assisted migration tool
- Claude
- with structured prompts, converting legacy logic into PySpark
- Tests included
- unit tests and Terraform generated with every conversion
A few practical questions
Questions about building with AI
What engineering and security leaders ask before the first change is made.
Ask us about your situationWhich AI tools do your engineers use?
Claude and other current models, matched to the task. Where code cannot leave your environment, we can run models privately or locally. The tools change quickly; the checks around them do not.
Does a person really review every change?
Yes. The review agent catches a lot, but it does not approve anything. An engineer reads each change with its test results, scan results and the review findings, and decides whether it ships.
Why use a second agent to review the work?
A model checking its own output tends to agree with itself. A separate agent, asked only to find what is wrong, catches problems the builder missed, so the engineer starts from a change that has already been attacked once.
Does our code leave our environment?
Only if you agree to it. We match the setup to your security rules, from approved cloud models to private or local models running inside your environment.
Will this make the project faster or cheaper?
We do not quote a multiplier. AI makes volume work practical that used to be too slow to take on, such as clearing a vulnerability backlog or converting a legacy ETL estate. What we commit to is the scope and the quality bar: tests passing, scans clean and an engineer’s approval on every change.
Related services
Where the method is used
AI-assisted modernization
Vulnerabilities closed, frameworks upgraded, codebases consolidated and legacy code converted.
Learn moreDatabricks consulting
Lakehouse architecture and legacy ETL migration with AI-assisted conversion and generated tests.
Learn moreAI on your data
Governed, private AI that answers questions from your own data, with the query behind every answer.
Learn moreAI development
Language model features built into the applications your people already use.
Learn more
Next step
What would you build with AI first?
Tell us about the codebase or the backlog. Here is what happens after you reach out.
A short call about the work
What it runs on, what is holding it back and what your security rules allow.
An engineer looks at what’s running
Scope comes from your systems as they are, before anything is priced.
You get a written proposal
Covering scope, team, approach and price.