Share
Embed
Transcript
28 slides
-
Building Software in the AI era our biggest (r)evolution yet?
© ZENIKA 2026 All rights reserved - Proprietary & confidential 1 -
Who we are Asyraaf Ahmad Senior Consultant at Zenika •
• • Tech Lead for AI transformation on a Govtech Project Teaching AI Driven Dev classes Fun fact: Has experience collecting mice brains © ZENIKA 2026 | All rights reserved - Proprietary & Confidential Michael Isvy Head of Engineering at Zenika • • • • Instructor for AI productivity courses Consulting and Audit on many projects 2 days a week: Computer Vision project for SG government Fun fact: started calisthenics one year ago. Never too old :) 2 -
Zenika • Software Engineering and Consulting • French company •
70+ people in Singapore, 600 people worldwide • Nowadays: mostly AI Transformations • Training, Consulting, Projects • Led by passion © ZENIKA 2026 | All rights reserved - Proprietary & Confidential 3 -
Show of hands • Who works in IT? (Software Engineering,
Product Management, Design…) • Who is a Software Engineer? • Which AI tool do you use? Claude, Copilot, Codex… © ZENIKA 2026 | All rights reserved - Proprietary & Confidential 4 -
How we used to build software in 2022 © ZENIKA
2026 | All rights reserved - Proprietary & Confidential 5 -
How we used to build software in 2022 © ZENIKA
2026 | All rights reserved - Proprietary & Confidential 6 -
From Product Manager to Business stakeholders © ZENIKA 2026 |
All rights reserved - Proprietary & Confidential 7 -
And we would call ourselves Software craftsmen/women © ZENIKA 2026
| All rights reserved - Proprietary & Confidential 8 -
We had our issues • Building Software was expensive •
There was never enough software engineers • Product Managers were never satisfied with the team’s capacity © ZENIKA 2026 | All rights reserved - Proprietary & Confidential 9 -
Still we lived in peace, until… © ZENIKA 2026 |
All rights reserved - Proprietary & Confidential 10 -
2023 to 2026 • 2023: we started using ChatGPT for
ShadowAI ◦ Copy/paste from ChatGPT into coding environment • 2024: AI came into the developer environment ◦ Much more precise • 2025: Claude Code by Anthropic • 2026: Huge model improvements © ZENIKA 2026 | All rights reserved - Proprietary & Confidential 11 -
Real world stories Happened on real projects! © ZENIKA 2026
All rights reserved - Proprietary & confidential 12 -
2023: Product Manager needs to validate an idea TODO: add
calendar on wall (2023) © ZENIKA 2026 | All rights reserved - Proprietary & Confidential 13 -
2026: Product Manager needs to validate an idea © ZENIKA
2026 | All rights reserved - Proprietary & Confidential Claude Design demo 14 -
2026: presenting prototype to a stakeholder • Before showing your
prototype, we state expectations that a prototype is just a mockup • No backend, no database, etc • It is just part of the ideation process © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
Engineer's reality check • A mockup is not production software.
• Auth, edge cases and security still take months to build. • AI builds the facade; engineers strengthen the foundation. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
A Facade: Prototype vs Production • Prototype Facade (1 Hour):
Mock JSON data, unauthenticated routes, single browser execution, zero error recovery, zero test coverage. • Production Foundation (Weeks): Relational schema migrations, OIDC / RBAC, concurrency safety, audit logging, OpenTelemetry, regression tests. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
How we code in 2026 • AI agents scaffold the
implementation. • Most boilerplate is generated, but human engineers direct the architecture. • We spend less time typing code and more time reading it. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
From writing code to reviewing code • Writing code got
cheap; verifying code got critical. • Senior engineers move fast because they spot mistakes quickly. • Juniors face a steeper climb to build that intuition. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
The Junior Developer Dilemma • Senior Advantage: 10+ years of
debugging gives seniors instant intuition during code review. • Junior Crisis: Junior engineers historically built intuition by writing mundane boilerplate. • The Apprenticeship Gap: If agents generate all boilerplate, the junior learning pathway breaks. • The Organizational Fix: Train junior engineers on test construction, system architecture, and failure diagnosis. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
Everybody is a product builder • PMs build working prototypes
instead of writing long documents. • The boundary between designer and product manager is blurring. • Teams test real user flows in days instead of quarters. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
We still have bugs • AI writes code with confidence,
including edge-case bugs. • We use AI to generate test suites and catch regressions early. • Automated tests are now the safety net for generated code. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
Demo takeaways • Faster coding creates feature inflation: Jevons paradox;
we build things just because we can. • Product managers become the bottleneck when specs are unclear. • Teams need structured code generation, not blind acceptance of generated content. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
Demo takeaways • Jevons Paradox in Code: Cheaper code generation
increases overall demand for software, risking feature bloat. • Feature Inflation: Organizations build unneeded features simply because code was cheap to generate. • The New Bottleneck: When engineers build in days, vague product specs become the blocker. • High-Velocity Teams: Require disciplined product boundaries and executable specifications. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
What's next: Privacy and local models • Enterprise data privacy
is back in focus. • Running open models on-premise works today. • It protects company data, but requires real hardware budgets. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
What's next: Privacy and local models • Data Governance: Regulated
public sector workloads cannot send citizen data to external commercial endpoints. • Open Weight Models: Models running on internal GPU clusters match commercial APIs on core coding tasks. • The Trade-Off: Public cloud APIs offer zero setup; sovereign private clusters require hardware budgets and MLOps maintenance. © ZENIKA 2026 | All rights reserved - Proprietary & Confidential -
We are hiring!!! Software Engineers (Java/Python/Golang) Site Reliability Engineer Data
Engineers Platform Engineer Full Stack Developers Business Analysts (DevOps) © ZENIKA 2026 | All rights reserved - Proprietary & Confidential 27 -
Thank you for attending! © ZENIKA 2026 All rights reserved
- Proprietary & confidential 28