New Wave of Intelligent Automation, AI Christmas, and Shopify's 10,000 Remote Engineers
AI-driven Intelligent Automation is replacing RPA. Google and OpenAI lead the AI race, Shopify innovates remote work with GenAI, and the tech world pushes toward AI readiness by 2025.
The Rise of Intelligent Automation and Sunset of Traditional RPA đ¤
Recently, weâve explored Agentic Systems and their transformative impact on labor. These systems are revolutionizing labor by automating routine processes, driving business efficiency, and, hopefully, bringing the joy back to peopleâs lives. â¨
A similar mission led to the rise of Robotic Process Automation (RPA) companies nearly 20 years ago. For instance, UiPath, founded in 2005, became a pioneer in automating repetitive tasks. By the late 2010s, more general-purpose RPA platforms (like WorkFusion) had already begun integrating AI into their products. Soon after, industry-specific intelligent automation platforms emerged, such as our local Amsterdam-based unicorn, Datasnipper, focusing on the financial services sector. đ
GenAI and AI Agentic Systems are transforming the RPA market, introducing a new era for Intelligent Automation. Recent announcements from leading GenAI companies are just confirming this direction:
Anthropicâs Claude-powered Computer Use
Desktop App for ChatGPT
Similarly to the general-purpose RPA platforms, these general-purpose agentic tools empower business users (and end consumers) to automate their routine processes in a modern way.
Industry-Specific Intelligent Automation đ
While general-purpose AI tools are becoming more powerful, industry-specific intelligent automation solutions still thrive, especially in highly regulated and complex verticals. Those primarily focus on data intake and transforming unstructured data into structured formats.
Areas like accounting or legal see the most of a surge in intelligent automation startups. Professionals in these industries must shift from âdoersâ to âreviewers,â as out-of-the-box agentic systems take on most tasks. đľď¸ââď¸
Are we surprised many disruptive solutions are not coming from the established accounting or law giants? Misaligned incentives (e.g., time-and-materials pricing models) and the risk-averse nature of these firms, especially considering the non-deterministic nature of Large Language Models (LLMs), create barriers to innovate.
Opportunities for Automation and Challenges Ahead
Many verticals still have labor-intensive, inefficient, and routine processes ready for automation. The U.S. Bureau of Labor Statistics (likely will experience high load and peak traffic in recent months đ) is a great resource to explore those automation opportunities.
We anticipate the rise of hyper-verticalized intelligent automation startups targeting niche sub-industries. However, a wide variety of solutions makes it harder for enterprises to adapt to this fast-changing environment.
This creates a need for specialized agentic system service platforms and consulting companies that:
Connect the dots between the solutions đ
Integrate intelligent automation toolset into the enterprise systems
Prepare existing enterprise data to be utilized in such use cases
Straight-forward time-and-materials pricing models and elevated risk aversion prevent established companies in professional services, such as IT consulting or SI, from adopting outcome-based options and driving self-disruption.
While agentic systems and the next wave of intelligent automation promise huge benefits, one foundational issue remains. Success hinges on enterprises' ability to access and utilize high-quality data, a challenge that has existed since the advent of RPA and will drastically slow down the adoption of the new tools.
Big Week For AI
The past several weeks have been full of AI announcements; hence, we decided to dive deeper into some of them if you were not following or missed some important bits.
Google Almost Outshined OpenAI â¨
Google is clearly back in the game, recently announcing its âreasoningâ model. Gemini 2.0 Flash, with its native tool support and multimodal API, represents an important step toward agentic solutions.
During the announcement, Google also introduced four agents that use Gemini 2.0 Flash to use tools, call functions, and respond to the API in real time. The 32k token context window is the obvious downside of the new model compared to others.
Another major announcement from Google was Veo 2, their most advanced video generation model. Veo2 can already generate 4K videos, while Sora maxes out at 1080p. It also allows for creating videos that are up to 2 minutes in duration compared to Sora, which has a 20-second limit. The official blog has stunning video samples, check it out.Â
The model is available via controlled waitlists, and early testers prefer Veo2 over Sora.
In addition to the fantastic results with the video model, Google introduced Gemini Flash with Multimodal API just a day before Open AI announced its Advanced Voice with Vision. So, Google was better prepared for OpenAI's 12 days of releases than OpenAI itselfâwe felt that way just before OpenAI announced its O3 model.
All Hail the King: OpenAI and Their 12 Days of Releases
We already mentioned some of the key announcements in the previous edition â nothing outstanding until the moment it was!
Before we get to that part, let's get the full context about what was announced:
The o1 model is available for everyone. It is faster, smarter, and supports images. o1 and o1-mini can now be fine-tuned, and the o1 API also supports image input
Pro Subscription for âŹ230 per month gives unlimited access to o1/o1-mini/GPT-4o, Advanced Voice Mode, and o1 Pro
Advanced Voice Mode supports video streaming, and API now supports WebRTC
Sora has been released and is available under sora.com
Update to ChatGTP Canvas: real-time text and code editing interface, support for custom GPT models, and Python code execution
"Projects" introduced to ChatGPT: upload documents, custom instructions, chat organization
The â12 days of releasesâ didn't bring anything exhilarating except for minor improvements or already-expected releases (Sora and o1). But that was just before the last day when o3 and o3-mini were announced (in Public Safety Tests for now).
According to the early tests and benchmarks, o3 achieves results at the level of a Ph.D., surpassing humans in the ARC benchmark (~87.5%).
The ARC benchmark was used for long time to prove that models can't think like humans and cannot solve complex problems. OpenAI, with the o3 release, basically proved that the model can solve complex human-level problems. Some critics point out that it is just one benchmark, and there are discussions about training data sets. Still, this is HUGE.Â
Although the cost per task for o3 is HUGE as well.
AI Dominates Thoughtworksâ 31st Edition of Technology Radar
The latest release of Thoughtworkâs Technology Radar is a clear indication of AI-rush. For those unaware, The Radar is a well-known and very reputable report outlining emerging software development trends and technologies, categorized by their recommended adoption level (Adopt, Trial, Assess, Hold) based on Thoughtworks' experience and research.
Now, in the last one, AI is everywhere: tools, platforms, techniques, agents, RAG, eval, LLM observability, and more.
Some radar selections are debatable, though. We think Cursor and Raycast must be moved from the âTrialâ section to the âAdoptâ section. AI is all about fast adoption and fail-fixing things, no one has a recipe yet.
Shopifyâs 10,000 Remote Engineers Thrive On Pair Programming
Farhan Thawar, Shopifyâs head of engineering, shared the story with a pragmatic engineer about how they built a Black Friday Live Globe. That might be more of a gimmick, but Farhan also chatted with Lenny Rachitsky and shed some light on how Shopify runs engineering, having more than 10,000 engineers, who, by the way, ALL work remotely (well⌠95% do).
One of the interesting takeaways is perhaps Farhanâs approach to GenAI tooling. Heâs less excited that, for example, copilot will help engineers write code and more enthusiastic that it will actually help write less code or even delete lines of code. Considering that pair programming is somewhat of a holy grail of engineering practices at Shopify, copilots will contribute to this practice even further.
What can other companies learn from it? Remote teams must maintain intense collaboration among human contributors while leveraging GenAI's potential instead of replacing sparring partners with copilots.
đ Things Worth Checking
Andrew Ng gives a keynote on The Rise Of AI Agents And Agentic Reasoning
An interview with the first Neuralink patient brought a dose of techno-optimism: Despite being paralyzed from the shoulders down, Arbaugh regained a slight movement in his arm and hand. Neuralink also gave him the superpower to control a computer cursor with his mind.
Fantastic interview with Satya Nadella about his CEO journey, view on the AI future, and the company's vision for staying competitive across enterprise and consumer markets.
Sequoia claims that we are AI-ready for 2025 with all building blocks in place: foundational models, data, enough compute power, etc.
In other news, Nissan, after several failed attempts at EVs and under pressure from Chinese EV manufacturers, is on the brink of bankruptcy and is considering a merger with Honda.