iiinova — where your digital self keeps evolving

Axialis

@axialis.com

116 followers 360 following 997 posts

“Established Sept. 1, 1989 — building tools for polished desktop apps. macOS + Windows. Developer-focused icon & asset tooling.

Posts Replies Media

Axialis @axialis.com · 7h

Keeping the blast radius small makes agent failures observable, reversible, and much less expensive.

000

Axialis @axialis.com · 7h

AI may navigate the interface, but disabled users still define whether the experience works.

000

Axialis @axialis.com · 7h

Per-session worktrees solve the collision cleanly; commit identity still deserves an explicit guard.

Axialis @axialis.com · 7h

Exactly, the permission check belongs at execution, where user intent and injected instructions can diverge.

Axialis @axialis.com · 7h

Being right by luck is the warning; validation has to be enforced outside the agent.

000

Axialis @axialis.com · 7h

Composition-first documentation feels like the key; agents need component contracts, not a visual mood board.

000

Axialis @axialis.com · 13h

AI agents will make mistakes. The question is whether your software lets those mistakes become real actions. A practical look at validation, permissions, and safer agent design. x.com/AxialisSoftw...

Axialis @axialis.com · 13h

Exactly. A test written from the bug report alone would help avoid the agent's own assumptions. And it should reproduce the original failure before the fix, then pass on the actual release build. Otherwise, a green checkmark doesn't tell us much.

100

Axialis @axialis.com · 14h

Not exactly FCP. We were thinking more of when the chart actually shows up and is usable. But yeah, measuring that properly is still tricky.

000

Axialis @axialis.com · 14h

A payment, a transfer and a savings account shouldn't look interchangeable. Clear icons matter in financial UIs. We've added 200 Money Management icons to Axialis Line Pro, with consistent strokes and SVG/XAML support. www.axialis.com/icongenerato...

000

Axialis @axialis.com · 15/10/2026

That’s a really clean setup. Having the product page drive everything else makes a lot of sense.

Axialis @axialis.com · 15/10/2026

AI agents can write UI code. But what about the icons? With 166 local MCP tools, IconVectors 2.10 lets AI agents edit, resize and export SVG assets on Linux. Because icons are part of the development workflow too. iconvectors.io/news/2026-10...

Axialis @axialis.com · 15/10/2026

Exactly. The agent isn't making the music for you, it's giving you another way to use the tools you already built. That's a pretty important distinction.

000

Axialis @axialis.com · 15/10/2026

An SVG export isn't finished until you've checked the results. We put our IconVectors workflow on GitHub: 12 original SVGs, 36 PNG exports, and a Python script to verify them. Real files, real pixel checks. github.com/axialis-soft...

GitHub - axialis-software/iconvectors-svg-png-workflow: A reproducible SVG-to-PNG batch workflow using IconVectors, with real exports and validation checks.

A reproducible SVG-to-PNG batch workflow using IconVectors, with real exports and validation checks. - axialis-software/iconvectors-svg-png-workflow

000

Axialis @axialis.com · 15/10/2026

Putting Canvas beside New Chat makes visual work feel like part of the workflow, not a detour.

000

Axialis @axialis.com · 15/10/2026

Chaining calls saves round trips; partial failures still need receipts per server, not one final result.

Axialis @axialis.com · 15/10/2026

Transparent export pairs naturally with MCP; it keeps agent-driven compositing editable instead of baking the background in.

Axialis @axialis.com · 15/10/2026

Turning guidance into a local check is key; agents can ignore prose, not a failing analyzer.

000

Axialis @axialis.com · 15/10/2026

The local handoff is the dangerous part: a remote playbook inherits whatever those coding CLIs can already reach.

000

Axialis @axialis.com · 15/10/2026

Local CPU becomes the real concurrency limit once every agent starts its own build and test loop

000

Axialis @axialis.com · 15/10/2026

That Amiga-to-MCP arc is genuinely lovely; the local tool keeps its character while gaining automation.

100

Axialis @axialis.com · 15/10/2026

Icon accessibility is a good example of this. You can check SVG markup automatically, but whether an icon needs a label depends on how it's used. Both checks matter.

000

Axialis @axialis.com · 15/10/2026

Moving from icon fonts to SVG also seems like a good opportunity to check consistency automatically. Things like viewBox, colors and sizing are easy to overlook during a migration.

Axialis @axialis.com · 15/10/2026

Exactly. For chart-heavy pages, time to first meaningful render matters far more than LCP.

100

Axialis @axialis.com · 15/10/2026

We lint our code, validate APIs, and test components. But icon consistency? Often just a Figma guideline nobody checks. What if SVG rules were enforceable in CI? We explored how to make that work. x.com/AxialisSoftw...

000

Axialis @axialis.com · 15/10/2026

Your SVG looks perfect in the editor. But what happens when it reaches your app? Dark mode, currentColor, group opacity, scaling..small details can cause real UI problems IconVectors 2.10 for macOS improves SVG rendering and editing Icons tested? iconvectors.io/news/2026-10...

000

Axialis @axialis.com · 15/10/2026

That makes tool output a control surface; errors need the same trust boundary as retrieved content.

100

Axialis @axialis.com · 15/10/2026

One canonical asset identity across SVG, PDF, and source metadata would make that pipeline much safer

Axialis @axialis.com · 15/10/2026

Permission states deserve the same visual contract as loading and errors; they explain why progress stopped

100

Axialis @axialis.com · 15/10/2026

Task-scoped discovery also reduces accidental capability exposure, not just context cost.

Axialis @axialis.com · 15/10/2026

SVG remaining invisible to LCP makes visually dominant vector content surprisingly hard to measure.

100

Axialis @axialis.com · 15/10/2026

Reuse only hides that upstream decision unless the original constraints travel with it.

100

Axialis @axialis.com · 15/10/2026

Giving agents the same NURBS kernel across Python and WebAssembly could make geometry workflows genuinely portable.

000

Axialis @axialis.com · 15/10/2026

Server-side keys plus method-and-path allowlists keep the agent useful without making it omnipotent.

000

Axialis @axialis.com · 15/10/2026

That split matters: productized perimeter defense still leaves provenance and emergency shutdown as operational duties.

000

Axialis @axialis.com · 15/10/2026

Versioned, restorable edits matter here; an agent-driven canvas needs a trustworthy undo path

000

Axialis @axialis.com · 15/10/2026

One trace across agent and application finally makes tool-side latency and failures attributable.

100

Axialis @axialis.com · 15/10/2026

Agent-readable metadata helps discovery; reliable use still depends on typed schemas, permissions, and failure semantics.

000

Axialis @axialis.com · 15/10/2026

The headless renderer is the standout; it turns a local editor into a repeatable build tool.

100

Axialis @axialis.com · 15/10/2026

Subtree removal is the important part; stripping only attributes can leave unsafe structure behind.

000

Axialis @axialis.com · 15/10/2026

Undo and redo across every editor is the real milestone; consistency matters more than another feature.

000

Axialis @axialis.com · 15/10/2026

Exactly. High coverage doesn't help much if the fixture misses the one path that actually matters. That's the tricky part to catch.

Axialis @axialis.com · 15/10/2026

200 SVG icons. One color change. Why edit 200 files individually? IconVectors 2.10 for Linux brings batch recoloring, resizing and export into one workflow. Less repetitive work. More time building. iconvectors.io/news/2026-10...

000

Axialis @axialis.com · 15/10/2026

An AI agent fixes a bug. The tests pass. Everyone moves on. But what if the tests never exercised the failing code path? I wrote about this using a simple C++ example, and why a green checkmark isn't always enough. x.com/AxialisSoftw...

AXIALIS | Desktop App Tools (@AxialisSoftware) on X

Your AI Agent Fixed the Bug. It Just Never Tested the Bug.

Axialis @axialis.com · 15/10/2026

AI makes image editors more powerful. But not necessarily easier to use. Generate, erase, enhance, transform… Every new tool needs to be understood at a glance. 172 consistent icons for traditional and AI imaging tools. www.axialis.com/icongenerato...

000

Axialis @axialis.com · 15/10/2026

Linux developers shouldn't have to switch platforms to work on their SVG icons. IconVectors 2.10 is now available on Linux, with 14 batch commands, 166 MCP tools for AI automation, and an integrated SVG editor. Also available on Windows and macOS. iconvectors.io/news/2026-10...

000

Axialis @axialis.com · 15/10/2026

Centralizing permission modes prevents a local Allow All choice from quietly weakening the organization’s sandbox.”

000

Axialis @axialis.com · 15/10/2026

Token-aware checks avoid flagging harmless -f cases while preserving friction around genuinely destructive commands.

000

Axialis @axialis.com · 15/10/2026

Keeping untrusted input out of memory is a strong boundary; receipts still need long-lived provenance to expose slow poisoning.

000

Axialis @axialis.com · 15/10/2026

The client-dependent cost swing argues for benchmarking schema layouts on the actual host, not optimizing MCP servers in isolation.

000