The promise of AI agents automating office work has been loud for years, but the practical question for any operations leader is where to start. Aidan Pratt, the founder of Autostep, has a simple answer: start by watching. His company’s desktop application observes how knowledge workers actually spend their time, identifies the repetitive and costly tasks that go unreported, and then recommends or builds AI agents to eliminate them [Perplexity Sonar Pro Brief, retrieved 2024]. It’s a methodical, almost forensic approach to process improvement, one that aims to replace guesswork with a quantified ledger of operational waste.
For Pratt, this is about building a missing scorecard. The platform’s core output is what he calls a “P&L for knowledge work,” a ranked list of tasks by their hidden cost and impact [Perplexity Sonar Pro Brief, retrieved 2024]. The company’s early proof point is a claim of having identified over $100,000 in operational waste within a single 10-person team [Extruct AI, 2025]. That number, while from an undisclosed customer, is the kind of concrete ROI figure that gets a procurement officer’s attention. The bet is that direct observation, not surveys or integration-heavy audits, will provide the most accurate map of where automation dollars should go first.
The wedge of direct observation
Autostep’s primary differentiation is its installation method and data collection. The product is a desktop app that requires no initial integrations with existing tools like Slack, Jira, or Salesforce [Perplexity Sonar Pro Brief, retrieved 2024]. It runs locally, observing user activity to surface bottlenecks, waiting periods, and administrative loops. This passive monitoring is designed to capture the reality of work, which often diverges from documented processes. The resulting data builds what the company terms “compounding context”,a queryable record of operations that managers can use to ask how work actually gets done [Perplexity Sonar Pro Brief, retrieved 2024].
The workflow is straightforward: observe, quantify, and prescribe. Once high-impact tasks are identified, Autostep generates recommendations. These can range from simple prompts and templates to full “auto-agentic” AI agents, workflow automations, or suggestions for process changes [Perplexity Sonar Pro Brief, retrieved 2024]. The platform positions itself as the missing layer that measures both human and AI agent ROI in concrete outcomes and cost, framing itself as essential for serious AI adoption [X.com, 2026].
A founder with a pattern for automation
Aidan Pratt is a solo founder whose background reads as a deliberate build-up to this problem. A Georgia Tech student studying Machine Learning and Computer Science, he has already compiled a resume focused on automation and operational efficiency [Y Combinator, retrieved 2025/2026]. His prior roles include working under the CRO at fintech lender Cherry Technologies to speed up loan approvals, improving financial definitions with long-range steering at OpenAI, and processing thousands of due diligence documents at Opto Investments [sfnet.com, retrieved 2026]. He also previously founded Default, a B2B sales automation startup [Shapes.inc, retrieved 2026].
This pattern of applying technical skill to streamline business processes is the through-line. Pratt has also been a fellow at Kleiner Perkins, 8VC, and Z Fellows, giving him early exposure to venture-scale thinking [Shapes.inc, retrieved 2026]. While the company is young, founded in 2025, its backing from Y Combinator and association with venture firm Neo signal investor confidence in both the thesis and the founder’s ability to execute [LinkedIn].
| Role | Name | Key Background |
|---|---|---|
| Founder & CEO | Aidan Pratt | Georgia Tech ML/CS; ex-OpenAI, Opto Investments; founded Default (sales automation); Kleiner Perkins/8VC Fellow [Y Combinator, retrieved 2025/2026] [sfnet.com, retrieved 2026] [Shapes.inc, retrieved 2026] |
| Team | Machine Learning Engineers | Not publicly named; described as building the core observation and agent technology [Extruct AI, 2025] |
The early traction and investor confidence
Quantifiable proof of waste reduction is Autostep’s primary traction signal. The $100,000 figure from a small team is a powerful anchor for sales conversations, though the specific customer and methodology behind it are not public [Extruct AI, 2025]. The company is in an active hiring phase, with open roles for engineers and designers listed on Y Combinator’s job board, indicating a build-out of the core team [Y Combinator, retrieved 2026].
Funding details remain undisclosed, but the pedigree of support is clear. Participation in Y Combinator provides seed capital, network, and product discipline. The explicit association with Neo, as noted on the company’s LinkedIn tagline, suggests a pre-seed or seed round from the firm known for backing technical founders [LinkedIn]. For a tool that sells into enterprises, this early validation from institutional investors helps mitigate the classic “who else is using this?” risk during initial sales cycles.
Where the wheels could come off
The model is ambitious, and its success hinges on navigating several material risks. The most immediate is the value proposition for the individual employee being monitored. Autostep must convincingly frame its observation as a tool for eliminating drudgery, not for surveillance-based performance management. Any perception of the latter could trigger internal resistance and complicate deployment.
- Privacy and trust. A desktop app that observes all activity operates in a sensitive zone. Clear data governance, explicit user consent protocols, and a strong narrative about employee benefit are non-negotiable for adoption.
- Actionable prescriptions. Identifying waste is one thing; generating reliable, secure automations is another. The quality and maintainability of the AI agents or workflow changes Autostep recommends will determine its ultimate utility and renewal rates.
- Competitive response. The space for workflow intelligence and automation is crowded. Established platforms like Atlassian or Notion could add similar observational layers, while RPA and process mining vendors may move upmarket.
The company’s answer to these challenges appears to be its integration-light, desktop-first approach, which reduces implementation friction, and its focus on building a “compounding context” layer that becomes more valuable over time [Perplexity Sonar Pro Brief, retrieved 2024].
The realistic competitive set
Autostep does not compete directly with any single tool. Instead, it intersects several categories, each with different champions. Its realistic competitive set is a mix of point solutions and platform features.
- Process mining & task mining. Vendors like Celonis or UiPath’s Process Mining offer similar waste identification but typically rely on log data from enterprise systems, not direct desktop observation. Autostep’s wedge is its lack of integration prerequisite.
- Workflow automation. Zapier, Make, and native automation in tools like Slack focus on connecting apps after you know what to automate. Autostep aims to tell you what to automate first.
- Knowledge management. Competitors like Guru and Helpjuice help document processes, but they don’t actively discover deviations or quantify the cost of not following them.
- Platform expansion. The most significant long-term threat may come from productivity suites like Notion or Atlassian adding lightweight activity dashboards that surface repetitive tasks, leveraging their existing install base.
Autostep’s ideal customer profile is a pragmatic operations leader or engineering manager in a tech-forward company of 50 to 500 people. This buyer has budget for productivity tools, is feeling pressure to justify AI spending, and is frustrated by the gap between how work is supposed to happen and how it actually does. They need a quantified business case to secure budget for automation, and they prefer a tool that delivers immediate insight without a six-month integration project. For them, Autostep is selling not just automation, but the confidence to automate the right thing first.
The next twelve months will be about proving that the initial waste-identification case studies can scale into repeatable, multi-six-figure ACV contracts. The key milestone to watch is the announcement of a named enterprise design partner,a public validation from a company that has moved from a pilot to a rolled-out deployment. If Autostep can land that reference customer and begin to show retention metrics, the thesis of observation-led automation will move from an intriguing bet to a measurable market.
Sources
- [Y Combinator, retrieved 2025/2026] Autostep company launch profile | https://www.ycombinator.com/companies/autostep
- [X.com, 2026] Aidan Pratt on Autostep as an AI adoption scorecard | https://x.com
- [sfnet.com, retrieved 2026] Aidan Pratt speaker biography | https://www.sfnet.com/detail-pages/speaker-title/aidan-pratt
- [Shapes.inc, retrieved 2026] AI Chat with Aidan Pratt - Tech Founder & VC Fellow | https://shapes.inc/aidanpratt
- [LinkedIn] Autostep company profile | https://www.linkedin.com/company/autostep-ai
- [Y Combinator, retrieved 2026] Autostep job listings | https://www.ycombinator.com/companies/autostep/jobs