Matt Shumer co-founded OthersideAI, the company behind HyperWrite. In his 2026 essay Something Big Is Happening, he described changes he saw in his software work and urged readers to try AI on tasks from their own jobs. — Something Big Is Happening.

Part 1: The AI Tipping Point
- On the COVID parallel: Shumer compares the early reaction to AI disruption to the period when many people dismissed COVID as overblown. The analogy expresses his concern about how slowly people can recognize a change already underway. — Something Big Is Happening.
- On the end of incrementalism: Shumer describes the February 2026 releases of GPT-5.3 Codex and Claude Opus 4.6 as a turning point in his own coding workflow. — Something Big Is Happening.
- On tracking the frontier: Shumer sees a gap between how people outside tech experience AI and what he has seen it do on software tasks. His experience is the basis for his warning about broader adoption. — Something Big Is Happening.
- On warning others: Shumer says he wrote to give people he cares about a chance to prepare for changes he had already begun to see in his own work. — Something Big Is Happening.
- On the lack of preparation: In his February 2026 essay, Shumer forecast a disorienting two to five years and urged readers to start experimenting early. — Something Big Is Happening.
Part 2: The Evolution of Agents
- On autonomous action: Shumer introduced HyperWrite’s Personal Assistant as a tool that could act in a user’s browser and gather information, extending what users could ask it to do beyond drafting text. — Personal Assistant Launch.
- On operating browsers: Shumer describes teaching a browser assistant a repetitive workflow by recording a demonstration for it to repeat. — Agent Trainer Announcement.
- On outcome-based building: Shumer describes spending more of his coding time specifying what he wants built and delegating implementation to an AI tool. — Something Big Is Happening.
Part 3: The Future of Cognitive Work
- On cognitive substitution: Shumer argues that AI could affect several kinds of computer-based work because the same general-purpose tools can be applied to tasks such as drafting documents and analyzing information. — Something Big Is Happening.
- On the end of skill moats: Shumer advises making experimentation a habit so that a change in tools does not leave you dependent on one workflow. — Something Big Is Happening.
- On financial resilience: Shumer advises readers worried about industry disruption to think about how their savings and fixed expenses affect their room to adapt. — Something Big Is Happening.
- On dropping ego: Shumer argues that professional experience should help people test where AI fits into their work. He points to a senior lawyer who actively experimented with it. — Something Big Is Happening.
- On vulnerability: Shumer expects AI to affect substantial parts of jobs done on a computer. He identifies relationships, physical presence and legal accountability as factors that may slow adoption. — Something Big Is Happening.
Part 4: Prompt Engineering & Open Source Tools
- On automating discovery: Shumer’s gpt-prompt-engineer generates candidate prompts from a task description and tests them on examples. — gpt-prompt-engineer README.
- On ELO ratings: The project ranks prompts using Elo ratings from paired comparisons. The ranking depends on the chosen tests and evaluator. — gpt-prompt-engineer README.
- On prompt experimentation: The project treats prompt writing as something to test against examples. Its optional logs record the test cases, settings and rankings so an experiment can be reviewed. — gpt-prompt-engineer README.
- On accessible tools: Shumer distributed gpt-llm-trainer through notebooks so readers could run the workflow from a published example. — gpt-llm-trainer README.
- On automated fine-tuning: In gpt-llm-trainer, a task description is used to generate synthetic examples, split the dataset and fine-tune a supported model. — gpt-llm-trainer README.
- On reinforcement learning: Shumer identifies AutoRL as a project he worked on with OpenPipe. — Matt Shumer’s About Page.
- On deep research agents: Shumer’s OpenDeepResearcher prototype generates search queries, fetches and assesses relevant pages, and synthesizes the material into a report. — OpenDeepResearcher README.
Part 5: The Reflection 70B Controversy
- On independent validation: The Hugging Face Open LLM Leaderboard published separate evaluation artifacts for a Glaive Reflection 70B checkpoint. The artifacts document that specific run and checkpoint. — Hugging Face Evaluation Artifacts.
- On the core concept: Shumer’s Reflection 70B model card describes a design in which the model can flag a possible reasoning error before producing its final answer. — Reflection 70B Model Card.
Part 6: Practical Advice for the AI Era
- On career insurance: Shumer suggests trying AI on a time-consuming task from your own work and refining the prompt when the first result falls short. — Something Big Is Happening.
- On the personal benchmark: Shumer recommends choosing a few job tasks you think current AI will struggle with and writing simple prompts to test them. — CNN Interview.
- On tracking capabilities: Shumer suggests rerunning your own work tests on newer models every few months. He distinguishes improvement on a task from the likelihood that a whole job will disappear. — CNN Interview.
- On early advantages: Shumer sees an early advantage in learning to use new tools on real work and showing colleagues where they help. — Something Big Is Happening.
- On daily usage: Shumer recommends setting aside regular time to try an unfamiliar tool or a harder task. — Something Big Is Happening.
- On expanding capacity: Shumer encourages people to try projects they previously put aside because they lacked time, money or a particular skill. — Something Big Is Happening.
Part 7: Vision for Software Development
- On the end of manual coding: Shumer says that, by February 2026, he could describe some software tasks in plain English and leave an AI tool to carry them out. — Something Big Is Happening.
- On trusting the output: Shumer recalls leaving an AI system to work on a request for several hours and returning to a result he considered finished. — Something Big Is Happening.
- Give the Reviewer a Fresh Context: Shumer’s Gauntlet Loop asks a fresh AI reviewer to compare the actual work with a quality benchmark without seeing the builder’s explanation. Use that comparison to identify the biggest gap, revise the work, and judge the new result again. — The Gauntlet Loop
- On iteration speed: Shumer reports that he has used AI to get a working first version of an app in about an hour. — Something Big Is Happening.
Part 8: Predictions on Models and Scale
- On developing judgment: Shumer says some decisions GPT-5.3 Codex made while working on his instructions felt to him like judgment or taste. — Something Big Is Happening.
- On rapid iteration: Shumer expects workflows built around today’s models to change as those models are replaced. — Something Big Is Happening.
- On recursive improvement: Shumer points to OpenAI’s account of using an early GPT-5.3 Codex version in its own development as an example of AI tools contributing to model development. — Something Big Is Happening.
- Give an Email Agent Its Own Inbox: Shumer built his email-driven assistant around a dedicated inbox. A separate address lets people assign work by email and gives the assistant a place to send results back. Shumer disclosed that he invests in the inbox provider he used. — The AI Employee Built on AgentMail