An individual task, even a hard engineering problem, is exactly the kind of thing an AI assistant is good at. Copilot does that well. So does Yesper. That's not where the difference is.
The difference is everything around the task. A deliverable like a traffic study, a bill of quantities or a requirements review is dozens of interdependent pieces that all have to stay consistent with each other and with the codes. Change one and the others have to follow. Keeping the whole thing consistent is still done by hand, and it's the part a task-by-task assistant doesn't touch.
Same models underneath. Copilot works on the task; Yesper works on the whole deliverable and holds it together.
Producing a deliverable runs like a relay race. Each piece finishes and hands off to the next, every step built on the one before. Using AI on each task speeds up the legs, but it's still a relay. Three problems come with that:
Everything waits on the critical path. Review is the final leg, after the work is already done.
An error introduced early travels all the way to the end before anyone catches it.
Change one input and everything built on it downstream is already based on the old version.
Instead of a relay, Yesper works on the whole deliverable at once. Each piece is checked against the standards as it's produced, work runs in parallel, and every change is carried through everywhere it matters. It comes down to three things:
Draft the documents, run the analysis, generate the output.
A deliverable has thousands of pieces that depend on each other. Yesper keeps them consistent with each other, the regulations, and current sources like Trafikverket and SGU.
Put the parts together into a finished deliverable, checked against your templates, the standards and codes, and built-in knowledge of what correct looks like.
This compares the everyday-assistant work both tools do, the part of Yesper that lines up against Copilot. The layer above it, producing the deliverable, is Yesper's alone.
Built into Word, Excel, Outlook and Teams. Strong for daily office work within M365.
Full AI assistant for conversation, documents and analysis — independent of M365.
Copilot is the in-Office assistant; the rest of the job needs Power Automate, Copilot Studio and Power Apps — licensed and set up separately.
Assistant, domain agents, workflows, deliverables and apps in a single product.
Copilot Studio can shape per-role agents with setup, but there's no shared organizational layer that learns across the team and compounds over time.
Configured per organization, project and role by business users — no development. Shared memory and templates build across the team and compound over time.
General-purpose AI, with no construction terminology or standard codes.
Understands AMA codes, technical standards and construction terminology — and expands searches to the right codes and synonyms automatically.
Azure-based with Microsoft enterprise security. Global infrastructure.
Azure, EU-resident. Shared, Private or Dedicated tiers. Entra ID for SSO.
Searches via Microsoft Graph. Limited format support, no page/row-level references.
Automatic indexing with full provenance chain: filename, page, row, document type. Support for all common file formats.
Can compare documents in Word. No systematic requirements review.
Systematic review of deliverables against requirement documents, standards and governing documents with full source tracing.
General image understanding in chat only — no blueprint/drawing interpretation, BoQ extraction, or Swedish-optimized OCR in document processing.
Vision AI for drawings + adaptive Swedish OCR (4-step fallback, virtually never fails). Reads IFC/DWG models to generate Bills of Quantities.
Agent Mode in Word/Excel/PPT. Iterative editing directly in documents. Multilingual.
Canvas mode for co-editing with AI. Version history, export to Word/PDF. Translation to any language directly.
No ready-made construction or Swedish-authority agents — each one is a custom build via Copilot Studio + external APIs.
Ready-made agents for Trafikverket, SGU, SMHI, Naturvårdsverket, Lantmäteriet and Allabolag, plus construction law and reference libraries.
Researcher and Analyst reasoning agents — multi-step research and live Python data analysis. General-purpose, not construction-specialized.
Dedicated agents for file management, search, comparison/analysis, tender comparison and planning. Orchestrator coordinates multiple agents automatically.
Native in Office apps with Agent Mode. Strong integration with brand kits and templates.
Generates all document types with full formatting. Organization templates applied automatically. Native PDF creation and form filling.
Excel formulas, Power Automate, and the Analyst agent's sandboxed Python for data analysis. Not general stateful code execution or automation of external tools.
Full stateful code execution (Python, JavaScript, etc.). Automate tasks in external tools (ArcGIS, Excel macros, etc.).
Copilot Cowork runs long, multi-tool tasks; Copilot Studio orchestrates connected and child agents. Spans multiple tools and is billed per task in Copilot Credits.
Built-in orchestrator with sequential/parallel execution, nested workflows and automatic synthesis of agent results.
Human-in-the-loop by default; scheduled and agentic runs via Tasks and Power Automate (separate setup).
Full flexibility per step: autonomous or manual review. Run on schedule, on triggers or manually — all from one interface.
No workflow builder in Copilot itself — that's Power Automate, a separate Power Platform product. Copilot's own agent runs are metered (e.g. 25 Researcher + Analyst queries per user/month).
Fully no-code visual builder with drag-and-drop. No limitations on number of workflows. Create in minutes.
Copilot Memory is user-level personalization only — tied to the individual. No shared business, project or workspace memory.
Two layers: workspace and project. Automatic learning. Extracts key facts and decisions. Full audit history and traceability.
Collaboration per app (Word, Teams). Templates per app, not unified across projects.
Shared workspaces for AI tasks. Organization library with templates that are shared and applied automatically.
Not in Copilot itself — that's Power Apps, a separate Power Platform product and license.
The code agent builds hosted apps — checklists, maps, calculators — shared in one click, no separate tools.
Microsoft branding. No white-label. Standard Microsoft help.
Full GUI customization (colors, logo, white-label). AI-powered help system — ask the assistant how a feature works.
Multi-model in M365 Copilot — Anthropic (Claude) and OpenAI under admin control. Anthropic models run outside Microsoft's EU Data Boundary (processed in the US; off by default for EU tenants).
Defaults to the latest models (Anthropic, OpenAI, Google, etc.), all running within the EU — including Claude via Google Cloud in Europe. Configurable per organization.