Skip to main content

AWS Adds Harness to Open Source SDK for Building AI Agents

Amazon Web Services (AWS) this week revealed it has added a harness to the open source software development kit (SDK) it makes available for building artificial intelligence (AI) agents.

First introduced last year, the Strands SDK makes it simpler for application developers to use AI models to build and deploy AI agents. The Strands harness provides a fully assembled, customizable AI agent that developers can use as a foundation for building agents designed to automate specific tasks.

Marc Brooker, vice president and a distinguished engineer at AWS, said the overall goal is to make it simpler for application developers to build an AI agent that has been trained to automate a task that is unique to their organization. In many cases, those AI agents are now being built and deployed by DevOps teams much like any other workload, he noted.

The challenge is that bringing together the right set of primitives required to build a custom AI agent can be difficult. The Strands harness provides access to all the shell, file and web tools in a way that AI models can reason across in addition to automatically loading any AI skills provided. It also enables long-term memory across runs and resumes an earlier conversation when given a session ID.

Additionally, the Strands harness delegates open-ended subtasks to a built-in helper agent along with a checklist for tracking multi-step workflows.

Just as importantly, AWS included a context window capability that offloads bulky tool results to files and caches that can be reused to reduce the total number of tokens that might be consumed by as much as 28%, said Brooker.

In general, AWS is clearly trying to make it simpler to build and deploy AI agents that run as efficiently as possible, noted Brooker. While AWS has no plans to donate the Strands SDK to a consortium, it does not restrict how its SDK might be used to build AI agents that could potentially run anywhere.

Mitch Ashley, vice president and practice lead for software lifecycle engineering at the Futurum Group, said, in effect, AWS is making it possible to create a working agent using one line of code. As it becomes easier to build AI agents, the focus of DevOps teams will turn more toward deployment, he added. That pace of deployment, however, will still be limited by what DevOps teams can observe, control, and, most crucially, prove that an AI agent has executed a task as required, noted Ashley.

While AI agents represent a new class of workloads that DevOps teams are being asked to manage, the same fundamental best practices will need to be applied. The issue is that organizations may soon have thousands of AI agents that not only have to be built and deployed, but also continuously updated. Most existing software engineering workflows were not designed with that level of scale in mind. As such, DevOps teams, if they have not done so already, should start preparing now for an era of agentic engineering that is only beginning to dawn.



from DevOps.com https://ift.tt/dAwyDKx
Labels:

Comments

Post a Comment

Popular posts from this blog

Exadel Records Strong Year with Surge in Client Roster, Additions to Executive Team and Record-Breaking Company Growth

Success comes from growing need for digital transformation solutions and services amidst the COVID-19 pandemic WALNUT CREEK, Calif., January 12, 2021 — Exadel (www.exadel.com), a global provider of digital engineering solutions and services, announces a successful 2020 including a burgeoning client portfolio, continued growth, including new executive team members and 2020 sales projections. This year, […] The post Exadel Records Strong Year with Surge in Client Roster, Additions to Executive Team and Record-Breaking Company Growth appeared first on DevOps.com . from DevOps.com https://ift.tt/2LMO6eg
Post a Comment
Read more

Rochelle Walensky on the Rocky Road to Normal

Rochelle Walensky on the Rocky Road to Normal By David Wallace-Wells from NYT Opinion https://ift.tt/0A5Wx6r internal-sub-only-nl, Coronavirus (2019-nCoV), Vaccination and Immunization, Rumors and Misinformation, Centers for Disease Control and Prevention, Walensky, Rochelle
Post a Comment
Read more

Security as Code is Becoming the New Baseline: Continuous Compliance in DevOps 

There was a time when compliance meant a quarterly ritual. Someone from security would walk over with a spreadsheet, ask a few questions, tick a few boxes and disappear until the next audit cycle. The infrastructure team would scramble to prove that yes, encryption was enabled, and no, that S3 bucket was not public anymore. Everyone felt relieved, went back to shipping features and quietly hoped nothing would drift before the next review.   That model is dead; it just hasn’t been buried yet.   The problem is not that teams lack security awareness. Most engineering organizations today understand that vulnerabilities need catching early and that production environments need hardening. The problem is that compliance has historically lived outside the delivery pipeline — treated as a checkpoint rather than a continuous practice. In a world where teams deploy dozens of...
Post a Comment
Read more