Back to Agents: Community update
Ever since Fetch.ai’s inception, agents and multi-agent systems have been one of our primary areas of focus. We always believed in agents’ potential as the next generation of computational paradigms. This is especially true in an era where technology is being increasingly integrated into our day-to-day activities, promising us great comfort but also creating an overabundance of information to the point that our lives feel more, and not less, chaotic.
On 21 August 2019, we made the initial commit to the Autonomous Economic Agents (AEA) Framework repository, kicking off a relentless development period. During this time, we launched the core AEA framework and its packaging system (skills, protocols, connections). We also published numerous individual packages and ledger plug-ins, allowing agents to interact with underlying blockchain layers. We worked on the original Open Economic Framework, and its sibling simple OEF (sOEF), enabling search and discovery for agents and their services. We launched the Agent Communication Network (ACN) as a custom peer-to-peer infrastructure that allows agents to communicate one-to-one, without any intermediaries, and launched the AEA Registry as a repository of agents and their components to encourage sharing and reuse.
From September 2021, after almost 2 years, we slowed the pace of development to let the dust settle, but also to shift our attention to the problem of application. For us, this was a period of uncertainty; we still needed to verify the applicability of the tech we came to love. However cool we thought our technology might be, if there was no demand for using it to solve real problems, development would not be sustainable. We wanted to explore how useful industries would find agent-based solutions, not only as a viable alternative, but the preferred way of solving their current problems. A related issue, though one that required just as much effort, was being faced with a set of viable application directions, and having to choose one to pursue given limited resources.
As part of this strategic exploration, we also examined paths to monetization for every player involved in an agent-based ecosystem, including us, the developers contributing to the ecosystem, and the industries using and applying the tech. We do not claim to have found all the answers, but enough that signaled to us the viability of agent-based systems as a technology with immense potential but also a workable path to realization.
Almost one year from when we slowed development, we came to the decision to recommit to the agent approach!
Upon a quick examination, it was clear to us that a general house-cleaning was needed in order to get rid of the dev. clutter accumulated as a byproduct of our overly accelerated development earlier on, but also of the subsequent period of shifted focus. Therefore, we dedicated most of Q4 2022 to cleaning up the AEA framework repository and improving the development experience. This includes:
- Switched from `pipenv` to `poetry` for managing dependencies in the AEA Framework project. `poetry` is a modern tool with an easy developer experience and unique features for managing dependencies (such as dependency groups) for which we added support in the project.
- Reviewed all dependencies, removed unwanted ones and created categories per use-case (e.g. docs changes, tests, package development, etc.) for lighter installations.
- Updated dependencies to their latest versions (where feasible) and ensured cohesion.
- Completely updated, and added the missing repository documentations (e.g., readme, contributing guide, development guide, code of conduct, and so on).
- Updated the makefile with an easier structure, command categories, added missing commands, and cleaned up outdated commands.
- Completely redone the CI/CD process and workflows to speed it up and make it modular. The CI/CD process was particularly affected by our rapid development cycle, having turned into a tangled mess that significantly slowed the pace of development (before the update, it would take around 3.5 hours even for small documentation changes!) There are still aspects which could be improved significantly, but we are in a better position than before.
During the year of reflection, we also took some time to collect, compile and reflect on the feedback we received on our agent-based technologies over the years, from our internal teams, industry partners and the larger development community. Perhaps the number one repeated comment, which we fully agree with, was about the difficult on-boarding and steep learning curve of the tech, especially for newcomers and non-experts.
For us, 2023 is shaping up to be “the year of agents” as we double-down our efforts on improving our agent-based offerings. In response to feedback, one of the key directions of focus will be improving the accessibility and usability of our tools; the ultimate objective being easy adoption of agent technology by developers and enterprises. This includes better educational and reference material, and improving the developer experience.
You can expect to see many announcements about existing products but also new features, products and initiatives not yet revealed. Be sure to follow our social media channels, so you are the first to know!
Twitter — https://twitter.com/Fetch_ai
Telegram — https://t.me/fetch_ai
Discord — https://discord.gg/WcgB7X2Kp9
LinkedIn — https://www.linkedin.com/company/27233415