LogiCast AWS News: Lambda Web Functions, Bedrock Cost Visibility, AWS Security Agent, and Forward Deployed Engineers

Season 5, Episode 36 of the LogiCast AWS News Podcast features hosts Karl Robinson, CEO and co-founder of Logicata, and Jon Goodall, Principal Cloud Engineer at Logicata, joined by returning guest Girish Mukim, AWS architect, AWS Community Builder, and AWS Ambassador. The episode covers four topics: a new flavour of AWS Lambda, improved cost visibility for Amazon Bedrock, AWS Security Agent’s published penetration testing results, and the surge in “forward deployed engineer” job postings across the tech industry.

iframe width=“560” height=“315” src=“https://www.youtube.com/embed/r9__lnQUfu4?si=Kub15VCsgpcEqSHO” title=“YouTube video player” frameborder=“0” allow=“accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share” referrerpolicy=“strict-origin-when-cross-origin” allowfullscreen>

AWS Lambda Web Functions

AWS recently announced Lambda Web Functions, introducing a second distinct flavour of Lambda alongside what are now called Event Functions. Jon described being caught off-guard by the announcement, having first heard about it from a keynote at a conference rather than through his usual news channels. Event Functions cover the traditional Lambda model - throw an event at it, something happens - while Web Functions are oriented toward web programming patterns, supporting frameworks like Express, Hono, Fastify, Next.js, and anything Node.js flavoured.

Jon acknowledged uncertainty about where Web Functions fit relative to existing services. “I’m struggling, if I’m honest, to understand the difference between a Lambda Web Function and a CloudFront Function because they seem to perform very, very similar tasks,” he said, noting that the announcement even describes Web Functions sitting as an origin behind a CloudFront distribution. His tentative read is that CloudFront Functions are very constrained - JavaScript only, very basic - while Web Functions offer more control and a broader range of supported languages.

One area where the distinction becomes clearer is the billing model. Lambda Web Functions charge per millisecond of active CPU use rather than per millisecond of total runtime, which means time spent waiting on a database or external call is not billed. Jon suggested this could make Web Functions particularly relevant for agentic APIs, where a function might spend considerable time reading and processing data without burning CPU cycles. Girish flagged another appealing characteristic - simplified multi-region deployment through a global endpoint flag - as a practical benefit for teams already working through multi-region architecture challenges.

Both agreed that adoption patterns and real-world case studies will be needed before the use cases become fully clear. As Karl put it, time will tell, and it is worth watching for examples of this being used at scale.

AWS Cost Explorer and Budgets Support for Amazon Bedrock Attributes

AWS has updated Cost Explorer, Budgets, and dashboards to support Amazon Bedrock product attributes, giving teams much more granular visibility into how Bedrock spending breaks down by model, provider, inference type, and token category - including the split between input and output tokens.

Jon described the previous state of Bedrock billing as frustratingly vague. “The bill up until this was just showing you Bedrock tokens - it’s sort of gone to Anthropic. OK, cool. It’s not said users or particular token counts,” he noted. While CloudWatch could provide some visibility, it required building custom dashboards, was eventually consistent, and forced finance teams to access parts of the console that engineers would rather keep restricted. The new attributes mean cost allocation tags can be applied to inference profiles, projects, workspaces, and IAM principals, making it possible to see which roles or applications are consuming the most tokens.

Jon highlighted input versus output token tracking as the most practically useful addition. With that data visible, teams can begin to assess whether prompt caching would reduce input token counts, whether setting max tokens would trim chatty outputs, and whether specific models are generating disproportionate costs. Karl framed this in terms of ROI: without project-level cost tracking, it is effectively impossible to assess whether any given AI initiative is delivering business value relative to what it costs to run.

Girish noted that AWS has historically been slow to bring Bedrock into its own management tooling - the pricing calculator was a previous example - but welcomed the addition, pointing specifically to the Budgets integration as a useful control mechanism. The ability to set a threshold and receive a notification when model usage crosses it adds a basic but important guardrail that teams have been asking for.

AWS Security Agent and the AWS Continuum Suite

AWS published a blog post detailing how its Security Agent - part of the AWS Continuum suite - performed against 37 real-world critical vulnerabilities in a controlled benchmark. The agent successfully exploited 35 of the 37, with the two exceptions reclassified as off-intended-path results rather than failures: in both cases the agent managed to capture the flag but via a different route than the one the benchmark was designed to test.

Jon found the methodology as interesting as the results. Because any public benchmark risks being present in a model’s training data, AWS had to build its own private capture-the-flag style harness. He noted this revealed a wider concern about agentic AI: “It’s absolutely optimized for the end goal, not necessarily the route that you want it to take, so you have to be really quite careful with that.” The median time to capture across the successful results was around four hours, covering a broad range of vulnerability types including SQL injection, cross-site scripting, remote code injection, and server-side request forgery. The only category where results fell short was remote code execution via file upload - in both cases the agent broke the target system rather than exploiting it cleanly.

Karl clarified that Continuum currently contains four features, though three are in some form of preview: penetration testing (the Security Agent discussed here), code scanning, code vulnerabilities, and threat modeling, with the last being in gated preview. Jon acknowledged he is not fully clear on the distinction between code scanning and code vulnerabilities, and both noted that the area overlaps with existing services like Amazon Inspector.

Girish welcomed the transparency of publishing the benchmark results, including the cases where things did not go as intended. He also raised a broader point: the same AI capabilities available to defenders are available to attackers, which is precisely why automated tooling of this kind needs to move quickly. Jon’s consistent position on Security Agent is that it does not replace a fully accredited penetration test, but it significantly lowers the barrier to running regular, reasonably capable security testing - which historically might happen once a year if at all.

AWS CEO Defends Data Centers Amid Public Backlash

An article on Trading View reported on AWS CEO Matt Garman speaking out in defence of data center expansion, as communities across the US and beyond push back against new facilities being built near them. The panel noted this is not a new phenomenon - it follows the same pattern as opposition to wind farms and solar installations, where people want the benefits of the output without the infrastructure nearby.

Jon raised a sceptical note about the survey underlying the article, which cited Fox News polling showing 71% of registered voters opposed to local data center development. He pointed out that the demographic Fox News reaches is specific enough that the number cannot be straightforwardly generalised. Girish expanded the picture beyond the US, noting that similar opposition has arisen in Canada and Finland, where a stay was issued on a data center development that would have used 400 acres of forest. His view is that economic arguments tend to win in the end, and data centers continue to be built - but the debate is not going away.

Karl and Jon both noted that many people simply do not connect their daily use of streaming services, smartphones, and online shopping to the infrastructure required to support it. The article also highlighted economic arguments in favour of data centers - tax revenue and local employment - though Girish was cautious about taking the numbers at face value without knowing more about the methodology. The conversation briefly touched on AWS’s sustainability pillar in the Well-Architected Framework, with Girish noting it is often the last section addressed in any review, and Karl arguing that genuine efficiency gains start at the code level rather than at the infrastructure layer.

The Rise of “Forward Deployed Engineers”

A CNBC article reported on AWS and Anthropic accelerating the hiring of what they are calling forward deployed engineers - specialists who embed with enterprise customers to help them actually build and use AI tools effectively. The broader context is that large enterprises have been adopting AI agents and then rolling them back when the expected returns do not materialise, creating pressure on AI vendors to demonstrate value more concretely.

Jon made no effort to hide his feelings about the job title. “Unless you are actively in a forward operating base in a combat zone, you don’t get to call yourself forward deployed,” he said, describing the appropriation of military terminology into tech roles as something he finds genuinely objectionable. Setting that aside, his read of the actual role is that it is a skilled engineer who can operate across technical and business-facing conversations and help customers solve hard problems - something that has existed under various titles for years. Girish pointed out that AWS, Anthropic, and OpenAI are each taking different approaches: AWS sends a small team to help build and then hand over, Anthropic focuses on training rather than hands-on delivery, and OpenAI is positioning itself as a longer-term delivery partner.

The article cited a rise in job postings for the role from fewer than 200 in 2023 to more than 7,300 in the first nine months of 2026. Karl questioned whether this represents genuinely new roles or is partly a rebrand of existing consultant and cloud engineer positions. The chart of top employers hiring for the role - led by Deloitte, Accenture, KPMG, and similar firms - reinforced the sense that this is largely consulting work under a new label.

---

The full episode is available on all major podcast platforms and on the Logicata YouTube channel. If you enjoyed it, the team would appreciate a rating or review.

*This is an AI generated piece of content, based on the LogiCast Podcast Season 5, Episode 36.*