Why Webtonic chose Amazon Bedrock for its AI-enabled SDLC
As a team at Webtonic, we have proven to ourselves, beyond a doubt, that AI in the Software Development Lifecycle (SDLC) is a game changer and that it is here to stay. So, with a multitude of choices, that are also changing at an immensely rapid rate, we needed to make sure we selected the best tooling that would work for us across all the capabilities in our SDLC.
We evaluated each of the options below against six criteria:
- Cost – Costs needed to scale with usage.
- Maintainability – We did not want to manage the underlying infrastructure.
- Agility – We needed to add new models as they became available and retire older models with confidence and ease.
- Security – Communications between our IDEs and the models had to be protected.
- IP protection – We needed to assure clients, with confidence, that their intellectual property, code and data would not be used to train the underlying models.
- Usage allocation – We needed to attribute usage to individuals, projects and departments.
We considered four additional approaches to AWS Bedrock, namely:
- Continue without AI in our SDLC
- Host GPU servers in AWS
- Build a cluster of Mac minis
- Use Claude Code and/or GitHub Copilot
Continuing without AI
Continuing without AI was not a realistic option. AI is rapidly changing software development and ignoring it would make us less competitive and irrelevant over time. Used responsibly, AI acts as a force multiplier, helping our team deliver more value to our clients.
Hosting GPU servers in AWS
We considered hosting our own GPU servers in AWS but the cost would have been difficult for a small business to sustain. This approach would have given us larger fixed costs every month, rather than the goal of costs mirroring usage. The additional infrastructure would also have increased our operational workload, further increasing the cost burden.
Building a Mac mini cluster
A cluster of Mac minis appealed to us as builders but it would not have scaled as models and agents required more processing power. We would also have been responsible for availability and uptime, creating even more operational overhead than self-hosting GPU servers in AWS.
Claude Code and GitHub Copilot
Claude Code and GitHub Copilot were our initial choices because they are easy for a team to adopt and use productively. We used them very successfully and completed some impressive projects. These projects using the AI tools proved the value, however, we would have needed to monitor provider changes continuously to ensure that they did not undermine our commitments to clients, especially around IP protection and cost control.
Why we chose Amazon Bedrock
As an AWS Advanced Tier Services Partner, Webtonic naturally included Amazon Bedrock in the evaluation. It met all six of our criteria: cost, maintainability, agility, security, IP protection and usage allocation.
Amazon Bedrock lets us set budgets, control AI spending and allocate usage to specific people and projects. Comprehensive logging provides an auditable record of every AI call. Most importantly, we can assure clients that their data remains protected as it moves between the IDE and Bedrock, and that their intellectual property, source code, and data are never used to train the underlying models.
The teams at Webtonic have been working on this initiative over the last few months and we are in a continual cycle of improving how we use AI on Bedrock in our SDLC. The evidence is proving that work, which would have taken days or weeks, is now being completed in hours. This allows us to focus on doing more analysis and making sure we are building the right features for our clients, while still maintaining architectural integrity of what we deliver.