A legal software organization says it roughly tripled R&D output per engineer over 18 months by redesigning how work moved from product planning through deployment—not simply by using AI to write code faster. Greg Ingino’s 2026 account in InfoWorld describes fewer handoffs, AI assistance across the lifecycle, and quality and security checks built into delivery. The results are organization-reported, not independently audited, and the published account omits data needed to reproduce them.
What changed: work moved across fewer handoffs
The central change in Greg Ingino’s account was organizational. Instead of passing a feature from product to development, then QA, security, and deployment operations as separate stages, teams took more end-to-end ownership. AI agents assisted at points along that lifecycle, but the article attributes the largest gains to removing the boundaries and waits between stages.
As an Amazon Associate I earn from qualifying purchases.
That distinction matters: faster code generation alone cannot eliminate time spent waiting for decisions, test environments, reviews, security checks, or release coordination. The approach described aimed to reduce those delays while keeping controls in the delivery process.
Free tools Windows power users keep installed
One-click scans. No signup required.
“We assumed most of the productivity gain would come from AI writing code faster. It didn’t. The biggest gains came from getting rid of the handoffs between stages.”
#1 Best Overall
— Greg Ingino, InfoWorld, 2026
How AI fit into the engineering lifecycle
Requirements and shared context
The organization used a shared repository for product and engineering knowledge, giving people and AI tools a common reference for requirements and implementation context. AI assistance extended upstream into requirements work; Ingino reports that work which had taken weeks could be completed in an afternoon. That is a reported experience, not a general turnaround guarantee.
Code, tests, and review
AI assistance became more common in pull requests, and the account says AI generated 99% of new tests, with more than 39,000 AI-developed tests in the suite. The tests were connected to code changes and product requirements. These figures describe the organization’s reported practice; they do not establish that every generated test was correct, useful, or sufficient without review.
Rank #2
Quality and security controls
Quality and security gates were embedded in delivery pipelines rather than left solely to late-stage handoffs. Confidence scoring routed some routine approvals automatically, while lower-confidence work went to people. The account does not specify the scoring method, thresholds, or which approval types qualified, so those details should not be assumed to transfer unchanged to another team.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Standardized tools and gradual expansion
The team standardized on a small set of coding agents. Ingino’s recommendation is to begin in one lifecycle area, learn from putting the workflow into practice, and expand from there. This makes the first deployment a chance to check how well an agent fits existing work and knowledge systems, how controls behave, and whether cycle time actually improves.
Rank #3
What the organization reported
All figures below are claims reported by Greg Ingino in InfoWorld in 2026. The article does not provide an independent audit, the underlying dataset, baseline values, detailed metric definitions, or team-size changes. They are evidence of what the organization says it achieved, not a controlled benchmark or a forecast for other engineering teams.
| Measure | Reported result |
|---|---|
| R&D output per engineer | Roughly tripled over 18 months |
| Releases | Nearly twice as many per quarter |
| Deployments | Increased from 82 to more than 155; the account does not specify the measurement interval alongside these figures |
| Customer-reported defects | Fell 65% per million lines of code over 18 months |
| AI assistance in pull requests | Rose from about 3% at the start to 68% at the time of writing |
| Vulnerability density | Fell 76% over the same period |
| New tests | AI generated 99% of new tests; the suite contained more than 39,000 AI-developed tests |
| Specification to working pull request | About four hours, compared with 15 days previously |
| Requirements work | Weeks previously; reportedly an afternoon after the change |
Ingino says the organization tracked DORA metrics, cycle time, pull requests merged per developer, and lines changed per developer against a fixed baseline. Because the account does not publish the baseline values or enough methodology to reconstruct the comparison, the reported tripling cannot be verified or cleanly compared with another company’s figures.
Throughput can move the bottleneck downstream
Engineering was not the only stage that needed to keep up. The account describes go-to-market readiness as a downstream constraint: documentation, enablement, customer-success briefings, and customer readiness had to advance alongside faster releases. Otherwise, a team can deliver software sooner without making the release easier for customers or internal teams to adopt.
For organizations borrowing this approach, release planning should therefore account for those downstream tasks rather than treating deployment as the finish line. The InfoWorld account does not quantify the time or staffing needed to solve this bottleneck.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to apply the lesson without assuming the same result
- Choose one lifecycle area. Start with a specific source of waiting or repeated handoffs, such as requirements clarification, test creation, or a routine approval. Avoid introducing agents across every stage at once.
- Give the workflow shared context. Organize the product and engineering information needed for the selected task so that both people and AI assistance can work from relevant, maintained material.
- Build checks into delivery. Define the quality and security gates that must hold, and decide which routine actions can be automated and which require human review.
- Measure end-to-end results. Establish a baseline and use consistent definitions for cycle time, release or deployment frequency, defects, security findings, and AI adoption. Track whether the whole feature moves faster, not just whether one task produces output more quickly.
- Check downstream readiness. Include documentation, enablement, customer-success communication, and customer readiness in the release workflow when they apply.
- Expand only after learning. Review results and failure cases in the first area, adjust the workflow, then decide whether the same approach fits another stage.
When comparing tools or workflows, useful questions include which lifecycle stages they cover, how they connect to existing work and knowledge systems, what quality and security controls they support, when they escalate to human review, and whether measured end-to-end cycle time improves. Also check whether the approach helps teams prepare releases for the people who must support and use them.
What the case study does—and does not—show
The case makes a practical argument for pairing AI assistance with workflow redesign, embedded controls, and attention to release readiness. It does not show that AI alone caused the reported gains, establish a universal formula for tripling output, or provide enough data to verify the size of the effect independently. Teams can take the operating principles seriously while treating the numerical results as one organization’s account.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
