From the Cisco edition of October 31, 2026
Axio Intelligence: Cisco’s 98% is a target, not a result; a four-week AI-rewrite proof should be scored on behavior equivalence, data checks and rollback
Axio Intelligence (software modernization consultancy) · Independent commentary on AI legacy-code modernization · United States (general guidance)
- Publisher
- Axio Intelligence
- Original publication
- Axio Intelligence blog, Oct. 30, 2026 (Brice Ayres)
- Source retrieved
- 2026-11-01
What happened
Writing the same day as the Fortune report, Axio says the 98% is a target, not a result, and that nothing has been reported from the four-week proof or about how rewritten code will be tested. It argues AI-generated code can compile, pass review and existing unit tests and still behave differently, because legacy behavior lives in edge cases, undocumented rules, integrations and data. It recommends scoring a proof on behavior-equivalence tests against production-derived golden-master data, coverage of changed modules, integration and data reconciliation and a rehearsed rollback, followed by staged cutover. It notes C++ undefined behavior matters most in embedded and networking software.
Why it matters
Directly useful to SLED IT leaders who are being asked “why aren’t we doing that?” about AI rewrites of their own legacy systems (benefits, licensing, student information and finance systems). The checklist is a practical acceptance framework for any state, local or campus AI modernization pilot, separate from Cisco’s own effort.
Evidence and measured results
Practitioner guidance and analysis of public reporting. No original data, case results or measured outcomes. The author’s firm sells modernization services.
Limitations and uncertainty
Commercial blog from a firm that sells modernization sprints (includes a call to action). Guidance is generic and not peer reviewed. It interprets, but has no inside knowledge of, Cisco’s project.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-01; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- SLED leaders face pressure to use AI to modernize legacy systems quickly, but lack a way to tell whether rewritten code behaves the same as the old code.
- Stakeholders
- Agency or campus CIO
- Application owners (benefits, finance, student systems)
- Program/business rule owners
- Data privacy officer
- Discovery questions
- Which legacy system slice is bounded enough for a four-week proof?
- Do you have production-like test data that can be masked?
- Who decides whether a behavior difference is a bug or a policy rule?
- Value hypothesis
- Bounded: a scored proof produces a defensible go/no-go and a realistic forecast before budget is committed.
- Potential engagement
- Fixed-scope four-week AI-modernization proof on one slice with an equivalence harness and rollback rehearsal.
- Unsupported claims
- That AI can rewrite a whole legacy estate in a set timeframe
- Any cost savings
- That Cisco’s approach has worked
Pre-sales engineering
Role takeaway
- Fit
- Applicable to any AI-assisted code modernization pilot; not tied to a Cisco product.
- Architecture and integration
- Strangler facade or API gateway in front of the slice; shadow traffic; parallel batch runs; data reconciliation against downstream systems.
- Prerequisites
- Masked golden-master dataset
- Staging environment with integration endpoints
- Characterization tests
- Constraints
- Personal and regulated data in test sets
- Integrations with state/federal partner systems that cannot change
- Security
- Mask resident and student data in golden masters; restrict access to staging copies.
- Proof of value
- Report equivalence match rate and list of differences
- Measure verification effort versus generation effort
- Demonstrate rollback in minutes without data loss
Delivery
Role takeaway
- Work
- Build equivalence harness, run the proof, document differences and decisions, rehearse rollback, report go/no-go.
- Dependencies
- Production data extracts
- Business rule owners’ time
- Staging environment
- Ownership
- Agency application owner with business program owner sign-off.
- Skills and adoption
- Testers and developers need characterization and contract-testing skills; reviewers need training on AI-generated change review.
- Governance checkpoints
- Privacy review of test data
- Business sign-off on each behavior difference
- Change board approval before any traffic shift
- Acceptance criteria
- Proposed: 100% of differences triaged with a named decision
- Proposed: rollback completed in under an agreed time in rehearsal
- Proposed: parallel run through one full business cycle with reconciled outputs
- Risks
- Hidden business rules deleted by refactoring
- Downstream integration breakage
- Underestimated verification effort
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Recommends a strangler facade or API gateway, shadow traffic, parallel batch runs over a full business cycle and stepwise traffic shifts with a live rollback switch.
Governance
Who approves, reviews and stays accountable for outcomes?
Human decision required on every behavior difference (bug versus hidden rule); recommends reporting match rates and verification effort to leadership rather than headline percentages.
Security and privacy
What data, permissions and controls need testing?
Golden-master datasets captured from production should be masked where they contain personal or regulated data, which is significant for SLED systems holding resident or student records.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Verification skills (characterization tests, contract tests, data reconciliation) and business-rule owners’ time are the main workforce needs for AI modernization pilots in SLED.
Procurement
What should contracts, pricing and exit terms secure?
Suggests buying a fixed-scope proof with a written go/no-go before committing to a full modernization contract, which suits public procurement phasing.
Operating model
Which teams own the service once it runs?
Staged slice-by-slice cutover with parallel runs and named decision-makers for each behavior difference.
Publication history
- 2026-10-31Cisco · Issue 035 resources
Stable resource ID: axio-intelligence-cisco-ai-rewrite-four-week-proof-must-test-behavior-equivalence-20261030