👨💻 Flutter GDE | 🌍 Founder @FlutterMENA | 🚀 Building apps & breaking limits | 🎤 Storyteller in tech | 🇧🇭 Bahraini | Tech with a human touch ✨
Joined July 2014
- Tweets 3,129
- Following 907
- Followers 1,154
- Likes 2,930
186 Photos and videos
Pinned Tweet
8 Dec 2021
I'm glad to announce that I'm officially a @GoogleDevExpert for @Dart and @FlutterDev
I am thankful for the amazing Flutter community and I look forward to contributing more to this community🎯
#flutter
21
10
119
Hot take: the most valuable feature in Antigravity 2.0
for Flutter teams is not the CLI, the MCP server,
or Agentic Hot Reload.
It's a Markdown file.
AGENTS.md scales with your team in a way no other
feature does. Every developer and every agent session
starts from the same rules.
Disagree? Tell me what you think is more impactful.
dev.to/sayed_ali_alkamel/ant…
57
Jun 13
At a bank, you cannot confuse activity with progress.
Regulators don't care how many tokens your team burned.
Customers don't care how many lines the AI generated.
They care if the feature works and the money is safe.
"More AI usage" is not a success metric. Shipped value is
25
Jun 13
Measuring developer productivity by tokens consumed is like measuring a surgeon's skill by how many incisions they make.
More cuts = busier surgeon = must be good.
The patient cares about one thing: did you fix the problem?
Token count is not the patient.
27
Jun 13
Everyone's installing Flutter Agent Skills. Almost nobody knows how to write a good one.
And a bad skill is worse than no skill: it eats your agent's context budget and fires on the wrong tasks.
So I read the official Flutter, Claude, Antigravity & Codex guides and turned them into one recipe. Full breakdown a copy-pasteable skill 👇
dev.to/sayed_ali_alkamel/how…
1
78
Jun 13
skillscore 0.2.0: score multiple skills in one command.
skillscore skill-a/ skill-b/ skill-c/
I tested it on three skills from @addyosmani agent-skills (52k stars) and it told me exactly where to look. 🧵
#AI #DeveloperTools #OpenSource
1
2
131
Jun 13
performance-optimization scored 77/C for two reasons:
1. It ships terminal commands but has no Safety section.
That is a -8 penalty from the Antigravity authoring guide.
2. The description has no action verb and no boundary clause.
That one is from the Anthropic guide. The agent cannot tell when the skill stops applying.
Both fixable in under 10 lines.
1
74
Jun 13
If you're writing skills for an agent, run it on yours and tell me where it's wrong.
I want the rubric to match how people actually author skills, so the findings you disagree with are the most useful signal I can get.
1
30
Jun 13
It's open source (Apache-2.0), written in Dart, and on pub.dev:
dart pub global activate skillscore
skillscore your-skill/
📦 pub.dev/packages/skillscore
💻 github.com/sayed3li97/skills…
1
29
Jun 13
I pointed it at the Flutter team's own skill, flutter-add-widget-test.
90/100. An A. It lost points only for never telling the model what *not* to do, and a description that doesn't say where the skill stops. Both real, both one-line fixes.
It runs offline. No LLM in the loop.
1
39
Jun 13
Here's the part most people miss: an agent keeps every skill's description in its context window permanently, so it can decide when to use it.
So a vague skill doesn't just go unused. It costs tokens on every prompt and occasionally fires on the wrong request. Bad skills have negative value.
1
13
Jun 12
"skillscore: a CLI that scores your AI agent's SKILL.md 0–100" by @Sayed3li97 #DEVCommunity dev.to/sayed_ali_alkamel/ski…
1
24
Jun 12
Agents are turning into a platform, and "skills" are becoming their unit of capability: a SKILL.md file an agent loads to learn a task.
But there's no quality bar for them yet. And a weak skill is invisible until it's quietly taxing every call.
So I built a linter for it. 🧵
#FlutterDev #Dart
1
47
Jun 12
Google released DiffusionGemma and it changes how text generation works at the hardware level.
Not faster prompting. Not a smaller model. A different architecture.
Here is what actually changed:
1
2
67
Jun 12
Instead of predicting token by token (GPT style), it starts with a 256-token block of noise and refines all tokens at once using bidirectional attention.
Every token sees every other token during generation. That is what enables self-correction — something autoregressive models literally cannot do.
Result: the bottleneck shifts from memory bandwidth to compute. On modern GPUs, compute is cheap. That is why you get 1,100 tokens/sec.
1
1
55
Jun 12
The catch: accuracy is lower than Gemma 4 on hard reasoning benchmarks (MMLU Pro: 77.6% vs 82.6%).
Best for: code infilling, real-time chat, inline editing. Not yet for: complex math, legal, multi-step reasoning.
Apache 2.0. 18 GB VRAM. Deploy with vLLM today.
For more:
dev.to/sayed_ali_alkamel/dif…
1
37
Jun 12
Jun 12
Google just shipped DiffusionGemma and the inference numbers are wild.
1,100 tokens/sec on H100. 700 on RTX 5090. Fits in 18 GB VRAM. Apache 2.0.
It generates 256-token blocks in parallel instead of one token at a time. Up to 4x faster than autoregressive.
The trade-off: accuracy is lower than Gemma 4 for hard reasoning. Speed vs quality — pick your workload.
Full breakdown:
dev.to/sayed_ali_alkamel/dif…
1
34
Jun 12
Google just shipped DiffusionGemma and the inference numbers are wild.
1,100 tokens/sec on H100. 700 on RTX 5090. Fits in 18 GB VRAM. Apache 2.0.
It generates 256-token blocks in parallel instead of one token at a time. Up to 4x faster than autoregressive.
The trade-off: accuracy is lower than Gemma 4 for hard reasoning. Speed vs quality — pick your workload.
Full breakdown:
dev.to/sayed_ali_alkamel/dif…
1
79
Jun 12
Your AI coding agent is a Flutter generalist.
It averages across years of Stack Overflow and hands you the most COMMON answer, not the CURRENT one.
Dart & Flutter Agent Skills close that gap. Official, free, and actually good.
A thread 🧵🥊
1
1
59
Jun 12
My verdict: substance, not hype.
Skills make your agent reliably correct on the easy 80%: layouts, routing, tests.
They won't make it inventive on novel architecture. Guardrails, not a brain transplant. Still a trade I take every day.
1
1
12
Jun 12
Full breakdown: a real skill's internals, the catalog, and how to write your own 👇
dev.to/sayed_ali_alkamel/flu…
Tell me which skill saves you the most. 🥊
1
10