
Building AI Chatbots since 2016 @Chatbotsa 🚀 https://teapawt.org/t.co/EKLwmvKY5K - SaaS organic growth on autopilot 🤖 https://teapawt.org/t.co/jcL5hrG9pV
Cairo, Egypt
Joined October 2011
- Tweets6.9K
- Following1.2K
- Followers621
- Likes375
Pinned Tweet
I have been building a few #MicroSAAS ideas over the past 2 years, but they all failed. So I went ahead and analyzed over 1,000 SAAS ideas to spot the success patterns in product-market fit, #growth , and other hidden factors.
-
I am starting to build again but this time with 1,000 SAAS experience
-
I will be sharing my journey here, subscribe to the newsletter "link in bio"
#TheSAASGame
2
5
785
1/ All 7 of Martina’s AI teammates are live 🎉
Let me tell you about one of them: Otto 📬
He does your cold email outreach, from your own mailbox, every day, on autopilot.
I am building Martina - your marketing cofounder as part of 100K hackathon by @shipordie_ 's @marclou & @jackfriks and @higgsfield_ai
Martina learns your business and runs seven teammates inside the Claude Code or Codex you already pay for.
check below 👇
1
6
6/ 10 emails a day by default, up to 30.
One follow-up after 5–9 days. Never emails the same person twice. Tracks replies and opt-outs.
Want to check everything first? Turn on draft mode and approve each email.
1
1
7/ Otto runs inside the Claude Code or Codex you already pay for. He replaces cold email tools like Instantly or Lemlist.
The other 6 cover Reddit, LinkedIn, social, video, SEO and launches.
From $12.42/mo → getmartina.com
1
I am building Martina - your marketing cofounder as part of 100K hackathon by @shipordie_ 's @marclou & @jackfriks and @higgsfield_ai
Martina learns your business and runs seven teammates inside the Claude Code or Codex you already pay for.
check below 👇
getmartina.com
Stop paying for five marketing tools. Martina learns your business and runs seven teammates for Reddit, email, LinkedIn, social, video, SEO, and launches every day, inside the Claude Code or Codex you already pay for.
8
Ambient mode is the eve 1.0 piece I'd look at first.
The agent can listen in channels you choose without being mentioned. To limit model costs, a small decision model screens every message, and only the ones that might be for the agent start a turn.
A cheap gate in front of the expensive loop, so "always listening" doesn't mean a full agent turn per message.
11
Agent proposes, the system that owns the record decides.
That's the write path Salesforce describes for Copilotforce: Copilot will be able to take Salesforce actions, but CRM writebacks need explicit user confirmation and still pass Salesforce's existing validation rules.
Introducing Copilotforce
Salesforce customer context and sales workflows are coming to @MSFTCopilot and @MicrosoftTeams—where more than 400M people already collaborate
Our expanded partnership with @Microsoft spans Copilot, Teams & Fabric:
→ Salesforce in Microsoft Copilot will help sellers prepare for meetings, spot deal risks & take supported Salesforce actions, informed by customer context, emails, meetings & shared documents
→ @SalesCloud in Teams will bring pipeline insights, deal alerts & supported opportunity updates into Teams, with Agentforce Coworker helping sellers delegate work & take action
→ Data 360 + Microsoft Fabric connect through bidirectional zero copy, so teams can access data across both platforms without duplicating source datasets
→ @Informatica expands its Microsoft integrations to help teams find trusted data in Copilot, build governed AI agents in Foundry & trace data lineage in Fabric
@Accenture joins as Customer Zero and design and implementation partner to help shape and validate Copilotforce within its global sales organization, after scaling Microsoft Copilot to 740,000+ people
See what’s new from #MSIgnite: sforce.co/4kgw4iZ
8
The detail I'd keep from this Anthropic piece:
On a sample support agent, folding one subagent into another cut latency but broke a process that had been passing. The main agent answered from the order lookup tool instead of asking the shipping specialist for the delivery date.
It only showed up in one of two runs. Run each eval task more than once before trusting a merge.
This by Anthropic seems both right & a condemnation of many “agentic solutions” & harnesses being built at companies. Rather than a sprawl of specialized subagents built by different teams, you often get better & cheaper results using one agent to do it all claude.com/resources/article…
5
Second GPT-6.2 migration detail, easy to miss:
In the Responses API, 6.2 appends top-level `instructions` after `input`. If your reusable system prompt lives there, OpenAI says to move it into a `developer` message as the first item in `input` so it stays eligible for prompt caching.
Keep `instructions` for one-off, per-request guidance.
1
1
7
The most useful line in this launch for anyone running agents isn't a benchmark.
OpenAI says that in two alignment tests, both models were more likely to omit mentioning obstacles that blocked their work, like a missing file or an unavailable search tool. They added Codex instructions that substantially reduce it.
In your own agent loop, make "what blocked you" a required part of the final report.
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
openai.com/index/introducing…
9
Two MCP items in Claude Code 2.1.327 worth knowing:
• `claude mcp add-json <name> -` now reads the server JSON from stdin, so tokens stay off the command line
• Fixed: `claude mcp add/remove` at project scope deleting other top-level keys in `.mcp.json`, like `$schema` or another tool's `servers`
If your repo shares `.mcp.json` with other tools, that second one matters.
1
11
لو الـ request بتاعك مش مستعجل، OpenAI نزّلت في الـ Responses API (public beta) حاجة اسمها completion window: تختار 2h أو 8h والـ token rates تقل 30% أو 40% عن الـ Standard.
والحلو إن الـ follow-up requests في نفس الـ agent run بتاخد نفس الخصم. يعني مراجعة documents بالليل أو تحليل tickets قبل اجتماع الصبح بقت أرخص.
بس خلي بالك: لازم Standard tier ومفيش streaming.
25
Anthropic shipped Linked billing (beta): your customers can link their own Claude org to your app and pay for the Claude usage it runs, with their own spend cap.
One line in the docs to read twice: requests still run in your org, so your data retention and keys apply, not the customer's ZDR or BAA.
1
15
GPT-6.2 Astra and Sol are in the OpenAI API now.
The migration note worth reading: from 6.2, the API drops reasoning items if you edit earlier conversation history, downgrade the model, or change the end user/org mid-conversation. OpenAI says quality can dip and token use can go up.
If your agent rewrites its own instructions between turns, check this one.
1
31
Summary of My Participation on Day 1 at the #ShipOrDie #100kHackathon
🔴 6:45 AM: I discussed the idea with Claude and used my [GitHub repository](github.com/eyoussef-builds/i…) to prepare everything.
🟢 10:25 AM: The project is fully set up! and live! I've connected Traffic and Stripe to the ShipOrDie platform.
📅 Tomorrow: I will test everything before starting the marketing phase.
1
46
Small coding-agent changes from this week worth knowing, straight from the official changelogs.
Claude Code 2.1.326: Bash commands that delete or overwrite files (find -delete, cat a > b, sort -o) were being summed up in the transcript as a search or a read. Now fixed.
If you skim transcripts to see what an agent touched, that label is what you were trusting.
3
59
GitHub's AI Scan for repositories can now be started on up to 20 repos at once (public preview).
It looks for vulnerabilities on the default branch, including languages CodeQL doesn't support, and findings land next to CodeQL alerts. No Copilot license required.
1
1
19






