February made it easier to imagine delegating substantial knowledge work to AI. It also supplied a reminder that jobs consist of many different tasks, with very different requirements. Research, software changes, and routine writing should not share a single success measure. This edition examines three developments through that operational lens: what became available, what was still experimental, and how a business could evaluate the opportunity without assuming a vendor's benchmark would become its own financial result.
1. Deep research offers a more substantial first pass
OpenAI introduced deep research on February 2 as a capability for multi-step internet investigation. It could search, analyze, and synthesize material into a report with citations. Availability expanded during the month, including an announcement on February 25 that Plus users could access it. The release presented a useful research tool, not an assurance that every report would be complete or correct.
A business owner could evaluate it on supplier comparisons, market background, or a competitor briefing. The best prompt would define the decision, geography, time period, and required evidence. Reviewing the bibliography is part of the task: a polished report may still miss an important source or merge facts from different periods. A useful result should separate what the source says from the analysis built on top of it. That separation matters especially when an attractive conclusion will influence spending.
Business takeaway: Commission one bounded research brief and have a knowledgeable reviewer audit its central claims. Measure total review time and decision usefulness, not just report length.
Source: https://openai.com/index/introducing-deep-research/
2. Claude 3.7 makes thinking time a design choice
Anthropic released Claude 3.7 Sonnet on February 24 with both standard responses and extended thinking. The API allowed developers to control a thinking budget. Alongside it, Claude Code arrived as a limited research preview for delegating coding tasks from a terminal. The model release and the coding tool's preview were distinct stages of availability.
For custom business software, this raised a useful design question: which tasks deserve more computation? A straightforward label or summary may not justify the same process as diagnosing a difficult defect. Evaluations should include response time, cost, and the quality of the final checked result. For code, the meaningful endpoint is a reviewed change that passes relevant checks in the application. A convincing explanation of a fix is not enough. The ability to inspect the change and reverse it remains valuable regardless of the model's claimed coding strength.
Business takeaway: Route simple and complex work differently. Trial the coding preview on an isolated, reversible change with an explicit acceptance test and an experienced reviewer.
Source: https://www.anthropic.com/news/claude-3-7-sonnet
3. An economic index focuses attention on tasks
On February 10, Anthropic introduced its Economic Index using anonymized conversations from Claude.ai. Its initial analysis found substantial use in software development and technical writing and classified more observed use as augmentation than automation. This was measured activity within one provider's product, not a representative census of every business or evidence of economy-wide job replacement.
The management lesson was to examine work at the task level. An employee may spend time gathering information, making a judgment, explaining a recommendation, and documenting a decision. AI can be useful in one part while being unreliable or inappropriate in another. Asking whether an entire role can be automated skips that distinction. A task inventory also makes costs more visible: time saved in drafting can be offset by additional checking or by poor information passed to the next colleague.
Business takeaway: Map the steps of one recurring process. Identify where AI assists, where it executes, and where a person remains accountable; evaluate those steps separately.
Source: https://www.anthropic.com/research/the-anthropic-economic-index
February's strongest business case was better-designed work. A research assistant, a reasoning model, and a coding preview could each contribute, but only when matched to a specific responsibility. Establish the baseline first, then ask whether the improved process produces a more accurate decision, a shorter turnaround, or a more reliable customer experience.
Present-day invitation: Book a conversation with Brice Prather at https://loyti.com/book to explore AI consulting, custom software, and business automation for your organization. Bring one operational bottleneck, the tools involved, and the business outcome you want to improve.
Brice Prather President, CEO & Founder Life On Your Terms, Inc. (LOYTI) Turning AI and custom software into smarter operations and business growth.