The Future of Work
The future of work is not a chatbox, it's an agent that can connect multiple tasks together across tools into reviewable deliverables and learn from feedback.
When you use ChatGPT, you type a question, get an answer, and copy what you need into your own document. Even as these tools get better context, using them still takes new habits, similar tasks still mean retyping similar prompts, and a finished draft still needs you to review, revise, and personally kick off the next step.
It’s useful, but a small slice of what’s actually possible. Consider how a simple highway sign gantry replacement project is created. A set of field photos is marked up with task comments in Bluebeam. Someone needs to resolve those comments and create a BIM sheet. That model is broken into a bill of work. Each task in that bill of work needs to have estimated costs referenced against a public pay-item catalog. This needs to be packaged into a final deliverable alongside a traffic control plan. And there are multiple rounds of revisions throughout, including a final quality control check.
Simply dropping AI into any individual step doesn’t yield material improvement: it just moves the bottleneck somewhere else.
Real transformation requires thoughtfully applying AI at each stage, connecting those stages together automatically instead of leaving a person to notice when the next task is ready to start, and keeping human review at the core of critical decisions.
Manual pre-2022 | AI Chatbots 2022–2025 | AI Background Agents sinusoidal, now | |
|---|---|---|---|
Starting a task | A person starts every task by hand | A person prompts the AI for each task | The next task starts on its own once a trigger fires |
Incorporating feedback | No way to reuse what was learned | Feedback disappears once the chat closes | Corrections get written into the process the AI follows, so the next run needs fewer fixes |
Picking the next step | A person decides what to do next | A person decides what to do next | One step's output automatically triggers the next step |
View some concrete examples
Suppose an agency needs to replace a sign gantry to bring it up to government code. An engineer works through the field photos, turning each marked-up comment into a detail on a CAD sheet, writes up the bill of work, and estimates the price for each task by referencing a public pay-item catalog.
Most of this is repetitive, mechanical work that requires attention to detail: defining tasks to align with standards and regulations, ensuring the comprehensive set of tasks are enumerated, estimating a realistic cost, and much more. None of it is hard, but a lot of that time goes into asking around, checking for best practices from someone who worked on a similar project, and essentially translating data from one source into a deliverable.
AI can do more here than speed up any single step: it can orchestrate the whole sequence, starting each task the moment its inputs are ready, and ground every step in the same living context, instead of a colleague’s memory.
- Drafting the sheet and writing the bill of work incorporates context from past gantry projects and current state regulations.
- Estimates cost by researching the relevant public pay-item catalog and past project costs on its own, compiling a supported number for a person to check with auditable methodology.
An employee still reviews key stages as well as the final deliverable. And with each review and edit, that tacit knowledge and newly completed deliverable can feed back into the shared ontology, so the next gantry project will go even smoother than this one did.
Employees go from being a “cog in the machine” to “architects of the machine” by managing Skills. Sinusoidal helps firms implement all of the orange boxes in the “after” state.
The mechanical parts of this work, translating a marked-up photo into a sheet, matching a bill of work against a pay-item catalog, get simpler, not just faster, so less of an engineer’s day goes to redoing the same lookups on every gantry.
That leverage runs in two directions: engineers get more room to learn where AI needs their judgment instead of picking it up ad hoc, and they can put more of their time toward higher-leverage projects that call for real design judgment and taste, now that AI handles so much of the mechanical work.
Where Sinusoidal helps
At Sinusoidal, we’re building the platform that makes that after state possible:
- Skills that get governed and continuously refined
- A compliant, cost-conscious, vendor-agnostic environment for the models to actually run in
- The interface your team uses to review work and collaborate with agents
And of course, none of that sticks without genuine buy-in from users, so we treat getting people to actually want to use it as part of the platform, not an afterthought.
We partner with firms on all of this work. If your team is looking for a place to start, we’d welcome that conversation.