DB1 Global Software
BRENES
Manifesto · AI First

AI changed how engineering gets done. We decided to lead the change.

This is our manifesto: how DB1 applies and governs artificial intelligence in building critical software. Forget the copilot bolted on top of the process. Here the development cycle was redesigned around an AI-accelerated method, with engineering governance in place of freestyle.

Specialized agentsHarness as governanceHuman gates26 years of engineering
The thesis

Most people use AI as a trick. We rebuilt the cycle around it.

A copilot here, a chatbot there, and the promise of productivity. For critical software that is not enough: it becomes speed without accountability. We believe in something else, and it fits in six sentences.
  1. 01

    AI without governance is technical debt at high speed.

  2. 02

    AI does not sign off. A human has their name on the gate.

  3. 03

    Specialized agents beat a generic model trying to do everything.

  4. 04

    Code in production is worth more than a report in PDF.

  5. 05

    26 years of engineering are not thrown away. They are taught to the machine.

  6. 06

    Speed without traceability is risk, not advantage.

The proof, on video

Legacy code modernization with 92% more performance

APIX 2026 event

Legacy code modernization with 92% more performance

The 92% that opens this manifesto is not a slide estimate. It is what our engineering presented on stage at APIX 2026, with the method open and the numbers on the table.

The concept

What Agentic Software Engineering is.

A development cycle in which specialized AI agents are a structural part of every step, not optional tools. Three things hold this model up.
01

Specialized agents

It is not a generic model answering questions. Each agent is specialized by method for a specific step: code diagnosis, business analysis, architecture, generation, testing and observability.

02

Harness as governance

A living artifact that encapsulates the rules: what is in scope, what goes in and out, the regulatory and quality limits, the acceptance criteria. It turns “generate code” into “generate code within these constraints”.

03

Humans in control

Validation gates with a chain of accountability. The AI proposes; the human approves. Whoever signs the gate has a name.

HARNESSguardrails active
Discovery
Specification▲ GATE
Implementation▲ GATE
Runtime
SAST · DAST▲ GATE
The agentic cycle

Discovery → Specification → Implementation → Runtime Intelligence.

01
Discovery

Understand the problem before touching the code. Business context, constraints and opportunities become a prioritized map of Intents.

Human gate · Validation
02
Specification

Define the Harness: boundary, contracts, guardrails and acceptance criteria that will govern what gets built.

Human gate · Harness
03
Implementation

CLOVIS generates code within the Harness constraints, with native tests and observability, validated by humans at every gate.

Human gate · Review + SAST/DAST
04
Runtime Intelligence

Monitor performance, usage and patterns in production. Anomalies and insights become new Intents and restart the cycle.

New Intents · continuous cycle
At every step, the gate is human. AI speeds up the path; whoever decides what goes to production has a name and answers for it.
CLOVISThe Harness is the contract. CLOVIS is what runs inside it. Our agentic implementation: it runs step 03 of the cycle, generates code within the constraints and stops at every human gate.
Below, one CLOVIS cycle: contract loaded, agents running and the human gate holding the merge.
db1 · clovis · agentic engineering LIVE
harnessGovernance
boundaries: core-banking, lgpd
guardrails: sast · dast · no-prod-write
acceptance: tests ≥ 90% · human review
agents.run4 agents
discovery.agentDiscoveryDONE
spec.agentHarness · SpecificationDONE
impl.agentImplementationRUNNING
runtime.agentRuntime IntelligenceWAITING
↳ impl.agent: generating tests · 17/24…1m12s
human.gatehuman control

review: tech lead · auditable trail · named approval

Approve mergeRequest changes
The difference

“Agentic Software Engineering” has a method behind the name.

Delivery
The market. The assessment becomes a report; implementation is left to another vendor.
DB1. From assessment to go-live, code in production with a single partner.
Use of AI
The market. Freestyle AI: raw output, no trail, “trust me”.
DB1. Agent → human code review → SAST/DAST → approval with a name on the gate.
Knowledge
The market. The generic model starts from zero in every conversation.
DB1. 26 years of patterns and pitfalls feeding the agents through a corporate RAG.
0.3% rework is not an accident, it is process. It is what happens when governance and AI go together.
In one table

Three ways to put AI in the code. One of them survives an audit.

Copilotin the IDE
Agentwithout governance
DB1Agentic Engineering
Auditable trail of what was generated
Human gate with the approver’s name
SAST and DAST before the merge
Boundaries per client context
Traceable back to the specification
Ready for regulatory audit
The proof

Method without results is opinion.

0%

Speed

faster than manual rewriting on complex legacy.

0.0%

Rework

rework, with engineering governance around the AI.

0years

Engineering

building and modernizing critical software, part of DB1 Group.

Audited by third partiesCMMI DEV/3ISO/IEC 27001ISO/IEC 27701ISO 9001

Who has been through this.

See all cases
The foundation

Our engineering opinion does not stay in-house. It is public.

Teaching 26 years of engineering to a machine requires having those 26 years written down somewhere. They are. Two open materials, in three languages, that anyone can read before talking to us.
01 · How we build

Engineering Guide

The DB1 way of building software, documented. Three tracks with practice, criteria and career progression: Engineering, QA and DevOps. It is what our teams follow and what our agents read.

Healthy software has a number
≥ 80%Healthy
60 to 80%Degraded
< 60%Sick

Coverage, maintainability, duplication, reliability and security. Measured, not estimated.

3 tracks · 36 documents · 3 languagesRead the Engineering Guide
02 · What we build with

Tech Radar

Our position on every technology that reaches production, across four quadrants and four rings. Nothing lands on the radar without having run in a real client project.

A sample of the rings
Adopt
Code ReviewCI/CDObservabilitySonarQubeKubernetesGitHub Copilot
Trial
JaegerAnsible
Assess
Claude CodeCodexDevinAgent SkillsAgents.MD
53 technologies · 4 quadrants · updated 03/2026Explore the Tech Radar

Claude Code and Devin sit in “Assess”, not in “Adopt”. It is the same criterion we apply to a new language. Your codebase deserves the same rigor.

Roberto Padilha, CTO of DB1 Global Software
Signed by

Roberto Padilha

CTO · DB1 Global Software

Technical chat

Bring your hard problem. We bring the engineering.

A conversation between technical people about your context: what is stuck today, what has already been tried and what Agentic Engineering changes in your case. No corporate deck.

  • You pick the slot on the calendar, with no email back and forth.
  • On the other side there is an engineer, not a sales script.
  • If it does not make sense for you, we say so on the spot.

DB1 Global Software · Agentic Software Engineering

Pick a time

Thirty minutes on our engineering team’s calendar. You get the invite with the call link right away.

  • 30 minutes. Straight to the point, no slide deck.
  • Your agenda. Legacy, architecture, AI governance, whatever is blocking you.
  • With the builders. Whoever joins the call is whoever signs the gate on your project.

Public calendar · no form

CMMI DEV/3ISO/IEC 27001ISO/IEC 27701