I run a OpenClaw AI agent called Obi. It handles my WhatsApp messages, journals notes, manages a CRM, does lead research. Throughout the day, an interesting information I want to deep dive later I screenshot and WhatsApp to Obi. It routes through one of a dozen LLM providers and notes it down in my journal.
Most days it works fine.
Normally, I forward an image from WhatsApp, a screenshot of an interesting tweet or a product page, and Obi would dutifully journal it. “Logged your note about AgentScope, Alibaba’s agent-centric LLM framework.” Sounded plausible. Except the screenshot was about Context Hub, a CLI tool for fetching API docs. Obi made up the entire description.
This happened four times in one session. Each time, the model received an image it couldn’t process (vision API rate limits were exhausted), got a [media attached: path.jpg] placeholder with no actual image content, and just… wrote something. Confidently. With details. A fabricated title, a fabricated description, saved to the database as fact.
I didn’t catch it for hours. The journal entries read like real notes.
We had a local PaddleOCR service running on the host machine, no rate limits, no cloud dependency. The fix seemed simple: route images through OCR first, get the text, then journal. We wrote the instructions into Obi’s memory file. Clear steps. Numbered. Imperative.
It didn’t work, but not for the reason you’d expect.
The OCR call takes 15-20 seconds. OpenClaw’s exec tool goes async after about 10 seconds and returns something like:
Command still running (session tidal-daisy, pid 76167).
To get the result, the model needs to call process(action=list), find the session by name, then call process(action=poll, sessionId="tidal-daisy") to get the output.
Here’s what actually happened, four times in a row:
| List showed | Model polled | Result |
|---|---|---|
tidal-daisy |
shiny-bay |
“No session found” |
crisp-tidepool |
warm-fern |
“No session found” |
briny-glade |
oak-mist |
“No session found” |
oceanic-pine |
violet-dune |
“No session found” |
The model would correctly call process(action=list), see the real session name in the response, and then immediately hallucinate a completely different two-word name when constructing the poll call. Every single time. Not a typo, not a partial match. A fully invented name in the same linguistic style.
After the poll failed, the model didn’t fall back to reading the result file. It just made up the image content and journaled it.
I tried the usual approach. Added instructions to MEMORY.md, which loads at session start. Added them again to HEARTBEAT.md, which gets injected every 15 minutes via cron. Used bold text, capital letters, numbered steps. “NEVER use process(action=poll).” “CRITICAL.” “DO NOT.”
This is what the industry calls SOP-based steering, structured operating procedures. Research from the Strands Agents SDK team at AWS shows SOPs achieve about 99.8% compliance across 600 evaluation runs, compared to 82.5% for simple prompt instructions.
99.8% sounds great until your agent runs hundreds of times a day. That 0.2% finds you. It found me four times in one afternoon.
I stopped trying to make the model follow instructions and started making the wrong action impossible.
The model can’t reliably copy a session ID between tool calls? Fine. Remove session IDs from the workflow entirely.
The new approach:
ocr_sync.sh with yieldMs=35000 (tells exec to wait up to 35 seconds before going async, long enough for most OCR calls to complete synchronously)process(action=list) only to check whether the job shows “completed,” not to extract a session ID/tmp/ocr_last.txt, a file the script always writes to, regardless of how the session tracking goesNo session ID ever needs to be copied, remembered, or passed between tool calls. The model reads a file. Files don’t get hallucinated.
After debugging this for a week, I wrote down what I’d learned. These aren’t theoretical. Each one maps to a specific failure I watched happen.
1. Models can’t copy arbitrary strings between tool calls. Session IDs, UUIDs, hashes, anything that looks like random words or characters will get hallucinated. Write results to predictable file paths. Use lookup-by-name instead of lookup-by-ID. If a tool parameter consistently gets hallucinated, remove that tool from the workflow.
2. Structured operating procedures beat prompt rules. A numbered checklist in a memory file that loads at session start is roughly 18 percentage points more reliable than the same instructions embedded in a system prompt. Keep the instructions imperative and sequential. Save conversational prose for humans.
3. Synchronous is safer than async. Every async operation introduces state the model has to track across tool calls. That state is where hallucination lives. If a command takes 15 seconds, set yieldMs high enough to let it finish synchronously. Only go async when you genuinely can’t wait.
4. Constrain, don’t instruct. When a model can’t reliably follow a “DO NOT” instruction, restructure the workflow so the prohibited action is no longer possible. Remove the tool. Change the API. Limit the inputs. “DO NOT” is a suggestion to a language model. A missing tool call is a guarantee.
5. Test at scale, not at demo. A workflow that passes 10 test runs will break in production. The long tail of unexpected inputs finds every gap in your instructions. Watch the first 5-10 real invocations after deploying any workflow change. Check logs for silent failures, cases where the right tool was called with wrong parameters.
The agent reliability problem isn’t about smarter models or better prompts. The models I was using are good. Qwen 3.5 122B is a capable model that handles complex tool chains well, most of the time. The problem is that “most of the time” isn’t good enough when your agent is writing to a database.
The fix was architectural, not linguistic. I didn’t write a better prompt. I removed the step that failed.
There’s a whole class of agent bugs that look like model problems but are actually workflow problems. Your model isn’t dumb. Your workflow is asking it to do something it can’t do reliably. Find that step, and replace it with something deterministic.
Prompts are suggestions. File paths are guarantees. Architect accordingly.
At oikyo.ai, we help organizations navigate the complexities of AI adoption, from strategy and platform selection to implementation and compliance. Whether you’re in Retail, finance, or any other regulated industry, we can help you build AI systems that are not just powerful, but trustworthy.
Contact us to learn how we can support your AI journey, or explore our services to see how we help organizations accelerate trusted AI adoption.
Interested in more AI insights? Subscribe to our newsletter or follow us on LinkedIn for the latest in responsible AI adoption.
]]>For decades, supercomputing was defined by “brute force”, throwing massive raw compute power at simulation problems. But as Nithin explained during our session, we are witnessing a paradigm shift. AI isn’t just running on supercomputers, it is fundamentally changing how supercomputing works.
Here are the four key ways AI is revolutionizing the field, along with a critical look at the risks of a widening technology divide.
The traditional scientific method involves testing thousands of hypotheses to find one that works. Nithin highlighted that AI is radically shortening this cycle by acting as a filter. Instead of simulating every possible scenario, AI models can predict and eliminate the less probable outcomes before the heavy computation begins. This is a game-changer for fields like vaccine development and drug discovery. By narrowing the search space, AI allows supercomputers to focus their immense power only on the most promising candidates, turning what used to be years of research into weeks.
Supercomputers are incredibly complex clusters of hardware, and managing their workload is an art form. We discussed how AI is being deployed to handle intelligent scheduling and workload balancing. AI agents can monitor the system in real-time, predicting bottlenecks and optimizing how jobs are distributed across the cluster. This ensures that every cycle of compute is used to its maximum efficiency, reducing energy waste and processing time.
One of the most practical applications we touched on was the ability of AI to spot “ghosts in the machine.” Hardware failures in supercomputing clusters can be catastrophic and expensive. AI-driven predictive maintenance monitors system telemetry to spot tiny anomalies—patterns in heat, vibration, or data throughput, that precede a failure. This allows engineers to fix issues before they break the system, saving millions of dollars in potential downtime.
Perhaps the most exciting frontier is Hybrid Intelligence. This is the pairing of AI’s pattern recognition capabilities with the raw calculation power of HPC to solve problems previously considered “computationally intractable.” This isn’t just about faster math; it’s about solving problems that were simply too complex for classical simulation alone. By combining these two intelligences, we are opening doors to new physics, climate modeling, and materials science that were locked just a few years ago.
Our conversation ended on a crucial, sobering note. As powerful as these tools are, they risk exacerbating the technology divide.
Much like the Industrial Revolution accelerated the wealth gap between industrialized and non-industrialized nations, the AI revolution threatens to leave behind those without access to this supercomputing infrastructure. If only a few nations or corporations hold the keys to these “engines of discovery”, the gap in healthcare, economic development, and scientific progress will widen drastically.
As Nithin and I discussed, equitable access is not just a buzzword; it is a necessity. We need to democratize access to supercomputing infrastructure to ensure that the benefits of this AI revolution—from curing diseases to solving climate change—are shared by the world, not just the privileged few.
This is where businesses, philanthrophists and influencers can come together to provide an effective approach to equitable AI, ensuring that access to the right supercomputing infrastructure is independent of you geograpic location and economic credibility.
AI for supercomputing has huge life-changing and world-changing potential, but we must keep in mind the need to ensure equitable access to these resources for everyone.
This article is based on an episode of AI Minute Mondays, where industry experts share insights on AI adoption, implementation, and impact across various domains. Watch the full conversation with Shish Shridhar above to dive deeper into the technical details and hear more about his journey in Retail and startups at Microsoft.
At oikyo.ai, we help organizations navigate the complexities of AI adoption, from strategy and platform selection to implementation and compliance. Whether you’re in Retail, finance, or any other regulated industry, we can help you build AI systems that are not just powerful, but trustworthy.
Contact us to learn how we can support your AI journey, or explore our services to see how we help organizations accelerate trusted AI adoption.
Interested in more AI insights? Subscribe to our newsletter or follow us on LinkedIn for the latest in responsible AI adoption.
]]>For years, we’ve had AI tools that assist shopping: recommending products, summarizing reviews, or helping us search. Agentic Commerce takes this a giant step further. Agentic Commerce is a model where AI agents are empowered to not just find what a user wants, but to autonomously go out, evaluate options, negotiate, and complete the entire purchase—from discovery to checkout—without human intervention until the final approval. As Shish pointed out, the shopper is no longer a person; it’s an intelligent, self-directed agent. “I have been thinking a lot about what happens when shoppers are no longer people but agents that negotiate, compare, and decide on our behalf. This new world of Agentic Commerce changes everything…” - Shish Shridhar
The rise of the autonomous agent demands a complete rethink of brand strategy and retail media. If an AI agent is making the purchasing decision, traditional advertising built on emotional storytelling and banner ads becomes less effective. The shift is profound:
Brands must pivot from trying to influence human clicks to providing the structured, machine-readable signals and trust frameworks that AI agents need to decide and recommend.
Product data is the new ad copy. Brands must optimize their content by providing transparent pricing, stock APIs, detailed ingredients, efficacy, and verified user data so the agent can accurately assess and compare offerings.
The old SEO playbook, focused on human search terms, is being replaced by Agentic Product Optimization (APO). This means optimizing for context—ensuring your product is the most logical, well-documented, and trustworthy choice when an agent searches for “a gentle moisturizer under $30 with SPF.”
This new commerce model necessitates massive changes in the underlying payments and digital identity infrastructure. The conversation highlighted a few key models that are emerging to facilitate agent-initiated payments:
The biggest remaining hurdle? Trust. Users must trust that the AI agent is acting faithfully on their behalf and that the transaction is secure and auditable.
For merchants, businesses, and platform builders, the time to start tinkering is now, while volumes are still low. The collective advice distilled from our chat and the broader industry landscape is clear:
Expose your product catalog, pricing, inventory, and policy data in structured formats (like Google’s structured product data or APIs) that agents can query directly.
Ensure your checkout flow can handle programmatic authorization, provide audit trails, and implement fraud tooling designed to understand agent behavior.
AI agents are not a threat to be blocked; they are the new customer. Develop systems like Stripe’s Model Context Protocol (MCP) or similar frameworks to let trusted, authenticated agents interact directly and effectively with your systems.
Agentic Commerce is not replacing human shopping overnight; instead, it will layer on top of existing digital commerce. For the next decade, retailers will need to operate in two modes: one for the human shopper and one for the agentic economy. The businesses that prepare to be data infrastructure providers and trust brokers for the agentic ecosystem today will be the ones that dominate tomorrow.
Agentic Commerce is poised to disrupt retail as profoundly as e-commerce did two decades ago. The winners will be those who balance innovation with responsibility, ensuring that AI agents empower—not alienate—consumers.
This article is based on an episode of AI Minute Mondays, where industry experts share insights on AI adoption, implementation, and impact across various domains. Watch the full conversation with Shish Shridhar above to dive deeper into the technical details and hear more about his journey in Retail and startups at Microsoft.
At oikyo.ai, we help organizations navigate the complexities of AI adoption, from strategy and platform selection to implementation and compliance. Whether you’re in Retail, finance, or any other regulated industry, we can help you build AI systems that are not just powerful, but trustworthy.
Contact us to learn how we can support your AI journey, or explore our services to see how we help organizations accelerate trusted AI adoption.
Interested in more AI insights? Subscribe to our newsletter or follow us on LinkedIn for the latest in responsible AI adoption.
]]>In a recent episode of “AI Minute Mondays,” Suchi sat down with Nishant Agrawal, Associate Director of Machine Learning at PathAI, to explore the transformative role of AI in pathology. Their conversation reveals not just the technical innovations, but the human-centered approach required to deploy AI responsibly in healthcare.
Nishant’s journey into health tech wasn’t traditional. With a computer science background and specialization in machine learning and computer vision, he found his passion at the intersection of technology and social impact. After studying at Carnegie Mellon, he joined PathAI—a Boston-based health tech startup that was pioneering the digital transformation of pathology.
“I don’t come from a typical biology background,” Nishant admits, “but after working there for seven-plus years, I could maybe look at a pathology image and tell you that this is a cancer cell versus maybe this isn’t. Don’t take my word for it, though!”
His story reflects a broader trend: technologists with deep expertise in AI and machine learning are increasingly finding meaningful applications in healthcare, where their skills can have life-changing impact.
Pathology has historically been a manual, analog field. When you undergo a biopsy or routine screening, tissue samples are stained with specific antibodies and examined under a microscope by highly trained pathologists. These experts zoom in, pan across vast tissue samples, identify patterns, and assign diagnoses—a process that’s both time-consuming and inherently subjective.
But the field is rapidly becoming digitized. Modern scanners can now generate gigapixel images of tissue samples, creating a fascinating computer vision problem ripe for AI-driven disruption.
PathAI focuses on two primary verticals where AI is making significant impact:
AI is helping biopharmaceutical companies accelerate drug development and identify which patients will respond best to specific therapies. A compelling example is PDL1 testing for immunotherapy.
The PDL1 Challenge:
The AI Solution: Machine learning can automate this quantification with greater consistency and accuracy. PathAI’s research shows that ML-powered analysis can identify more patients who would benefit from immunotherapy—particularly those near threshold levels (around 1% expression) who might otherwise be missed.
“It really isn’t the most comfortable feeling when you find out that this is something that pathologists have to kind of eyeball and assign a percentage to,” Nishant explains. “This thing really should be done by machines, and maybe some human in the loop can review it.”
In routine testing laboratories, hospitals, and academic medical centers, AI can:
These applications don’t replace pathologists—they augment their capabilities, allowing them to work more efficiently and focus on areas where human expertise is most valuable.
PathAI’s machine learning team consists primarily of ML engineers and scientists who build deep learning models and own entire product lifecycles. Their work involves:
“We use machine learning to build these products and own entire product lifecycles,” Nishant explains. “Working with our customers to understand what’s most painful and annoying—things we really shouldn’t be relying on humans to do if they’re not good at it—then building something well thought out, evaluating it robustly, putting it in the field, gathering feedback, and iterating again.”
Building AI for healthcare requires navigating complex challenges around regulatory compliance, ethical considerations, accuracy, and data bias. PathAI’s approach centers on several key principles:
Rather than establishing absolute “ground truth,” PathAI compares their AI systems against average pathologist performance on representative, intended-use populations. The FDA provides detailed guidelines on what these populations should look like, ensuring models are tested fairly and comprehensively.
“It’s never about letting the machine run amok,” Nishant emphasizes. “Our focus is always on the human and the AI in the loop—seeing how that can really transform the field and move it forward.”
This approach reimagines pathology workflows rather than simply automating existing processes.
PathAI produces heat maps that show pathologists exactly where the AI is focusing its attention. Instead of scanning entire slides looking for malignancy, pathologists can review AI-flagged regions and agree or disagree with the system’s assessment.
This interpretability builds trust and allows pathologists to understand the AI’s reasoning process.
Before deploying any product, PathAI conducts thorough risk assessments to identify potential scenarios and edge cases that could pose risks to patient outcomes. These assessments inform product design decisions that eliminate or minimize error potential.
Once products are in the field, PathAI continuously monitors their performance and proactively works to improve them based on real-world data and feedback.
AI in pathology represents more than just a technical upgrade—it’s a fundamental paradigm shift in how we approach diagnosis, drug development, and personalized medicine.
When implemented with proper controls and guardrails—robust evaluation, human-in-the-loop design, expert validation, and continuous monitoring—AI has the potential to:
For organizations considering AI adoption in healthcare or other high-stakes domains, PathAI’s approach offers valuable lessons:
As pathology continues its digital transformation, the opportunities for AI-driven innovation will only grow. Foundation models, advanced evaluation strategies, and novel computational biomarkers promise to unlock insights that were previously impossible to detect.
But success will require more than just technical sophistication. It will demand a commitment to responsible AI development, continuous collaboration with healthcare professionals, and a relentless focus on patient outcomes.
“It’s always fun and motivating to talk about machine learning and technology-adjacent fields in healthcare,” Nishant reflects, “and how we can help move the needle there.”
For technologists passionate about social impact, healthcare AI represents one of the most exciting and meaningful frontiers in artificial intelligence.
This article is based on an episode of AI Minute Mondays, where industry experts share insights on AI adoption, implementation, and impact across various domains. Watch the full conversation with Nishant Agrawal above to dive deeper into the technical details and hear more about his journey in health tech.
At oikyo.ai, we help organizations navigate the complexities of AI adoption—from strategy and platform selection to implementation and compliance. Whether you’re in healthcare, finance, or any other regulated industry, we can help you build AI systems that are not just powerful, but trustworthy.
Contact us to learn how we can support your AI journey, or explore our services to see how we help organizations accelerate trusted AI adoption.
Interested in more AI insights? Subscribe to our newsletter or follow us on LinkedIn for the latest in responsible AI adoption.
]]>Small Language Models (SLMs) are emerging as a practical alternative to their larger counterparts. While large language models (LLMs) like GPT-4 or Claude dominate headlines, SLMs are quietly revolutionizing how enterprises deploy AI.
1. Cost Efficiency
Running a large language model can be expensive. Inference costs, compute requirements, and energy consumption scale with model size. SLMs offer:
2. Domain Specialization
When fine-tuned for specific domains, SLMs can outperform larger models on targeted tasks:
3. Privacy and Control
Smaller models can run on-premise or in controlled environments, offering:
At oikyo, we’ve built our platform around the principle that the right-sized model, properly tuned, beats a generic large model every time. Our platform enables:
A 7B parameter model fine-tuned on medical literature can:
SLMs trained on financial data provide:
Domain-specific models excel at:
The future of AI isn’t just about building bigger models—it’s about building smarter, more efficient solutions. Small language models represent a pragmatic path forward for enterprises that need:
As the technology matures, we expect to see SLMs become the default choice for enterprise AI deployments, with large models reserved for truly complex, general-purpose applications.
Ready to explore how small language models can transform your business? Our platform makes it easy to:
Contact us to learn more about how oikyo can help you harness the power of small language models.
]]>Fine-tuning involves taking a pre-trained model and continuing its training on a specific dataset. This process allows you to:
The quality of your training data matters more than the quantity:
Do:
Don’t:
Choosing your base model is crucial:
Key hyperparameters to optimize:
Learning Rate
Batch Size
Epochs
Design robust evaluation:
Common techniques:
Compute Resources
Version Control
Our platform streamlines the fine-tuning process:
When fine-tuning causes the model to “forget” general knowledge:
Solution:
When test data influences training:
Solution:
Training a model that can’t run in production:
Solution:
Methods like LoRA and adapters allow:
Achieve good performance with limited data:
Keep models updated:
Define success metrics before starting:
Fine-tuning is a powerful technique that requires careful attention to data quality, model selection, training procedures, and evaluation strategies. By following these best practices, you can achieve excellent results while avoiding common pitfalls.
Ready to streamline your ML fine-tuning workflow? Try oikyo and experience the difference of a platform built specifically for fine-tuning from desktop to datacenter.
]]>Data scientists and ML engineers face a common pattern:
This disconnect leads to:
Historically, moving from development to production meant:
Infrastructure Changes
Process Overhead
Risk Factors
We believe AI deployment should be seamless. A model fine-tuned on your desktop should deploy to the datacenter with zero code changes.
Desktop Development
Datacenter Deployment
Docker containers and standardized environments ensure:
Start small, scale as needed:
Focus on models, not infrastructure:
# Fine-tune your model locally
from oikyo import FineTuner
model = FineTuner(
base_model="llama-3-8b",
dataset="./my_data.json",
config="./training_config.yaml"
)
model.train()
model.evaluate()
# Test locally before deployment
model.test(test_dataset="./test_data.json")
# Validate performance metrics
metrics = model.get_metrics()
print(f"Accuracy: {metrics.accuracy}")
print(f"Latency: {metrics.avg_latency}ms")
# Deploy to production - that's it!
model.deploy(
environment="production",
scaling="auto",
region="us-west-2"
)
No configuration changes. No code rewrites. Just deploy.
Deploy to any cloud provider:
Same workflow, different backends.
Intelligent scaling based on:
Built-in monitoring for:
Safety features:
Develop with production in mind:
Reduce manual steps:
Build resilient systems:
Track what matters:
A major bank used oikyo to deploy their fraud detection model:
Before oikyo:
After oikyo:
Results:
Ready to simplify your AI deployment workflow?
No complex setup. No infrastructure expertise required. Just seamless deployment from desktop to datacenter.
Get Started Today or Learn More about how oikyo can transform your AI deployment workflow.
Join thousands of data scientists and ML engineers who have simplified their deployment workflows with oikyo.
]]>Our conversation cut through the hype to focus on the reality of the threat landscape. While AI offers incredible defensive capabilities, it is also arming bad actors with sophisticated new weapons. Here is a breakdown of the top risks Neha identified and, more importantly, the roadmap she laid out for building resilient, secure AI systems.
During our chat, Neha highlighted that the barrier to entry for cybercrime is lowering. AI is allowing attackers to automate and personalize their campaigns at scale. Some of the areas of concern are listed below:
The days of easily spotted, typo-ridden phishing emails are fading. Neha warned of the rise of deepfake scams, where attackers use generative AI to clone voices and create hyper-realistic video avatars.
Perhaps even more concerning is the emergence of autonomous malware. Traditional malware often relies on a static set of instructions. AI-enhanced malware, however, can be “smart”.
Malicious actors can corrupt training datasets, leading AI models to make unsafe or biased decisions.
Carefully crafted inputs (like manipulated images or text) can trick AI systems into misclassifying or misinterpreting data, opening doors to exploitation.
Generative models can craft hyper-personalized phishing emails that are indistinguishable from legitimate communication.
Attackers can extract or replicate proprietary AI models, undermining intellectual property and security.
So, how do we innovate without leaving the back door open? Neha’s message was clear: Security cannot be an afterthought. It must be woven into the fabric of the AI development lifecycle.
“Garbage in, garbage out” is the old adage, but in cybersecurity, it’s “Vulnerability in, disaster out.”
You cannot wait for an attacker to find the cracks in your armor. You have to find them first.
Black-box AI is a security nightmare. If you don’t know how a decision was made, you can’t tell if the system has been compromised.
As Neha eloquently put it during our session, AI is not just the weapon, it is also the shield. By adopting a “secure by design, secure by development and secure by deployment” mindset, prioritizing governance, testing, and transparency, we can harness the full potential of AI while keeping our digital frontiers secure.
This article is based on an episode of AI Minute Mondays, where industry experts share insights on AI adoption, implementation, and impact across various domains. Watch the full conversation with Shish Shridhar above to dive deeper into the technical details and hear more about his journey in Retail and startups at Microsoft.
At oikyo.ai, we help organizations navigate the complexities of AI adoption, from strategy and platform selection to implementation and compliance. Whether you’re in Retail, finance, or any other regulated industry, we can help you build AI systems that are not just powerful, but trustworthy.
Contact us to learn how we can support your AI journey, or explore our services to see how we help organizations accelerate trusted AI adoption.
Interested in more AI insights? Subscribe to our newsletter or follow us on LinkedIn for the latest in responsible AI adoption.
]]>In my recent AI Minute Mondays episode with Tom Wisnowski, Principal Architect at Microsoft’s FastTrack Engineering Team, we explored a critical question: Is AI the pilot or the copilot in modern development? We explored where AI shines, where it struggles, and why the “Pilot vs. Copilot” distinction is the most important mental model a developer can have right now.
Tom’s perspective was clear: AI is your copilot, not your autopilot. It can assist, suggest and even refactor, but the human engineer remains accountable for the architecture, the code and ultimately the impact.
During our discussion, we highlighted several areas where AI delivers tangible value:
AI tools can generate scaffolding code and design patterns in minutes, reducing iteration cycles.
By surfacing edge-case bugs and suggesting fixes, AI helps engineers focus on higher-level problem solving.
Natural language models can auto-generate clear explanations, making systems easier to maintain and onboard.
AI can help product teams brainstorm and articulate user scenarios, ensuring coverage of diverse workflows.
By analyzing requirements and code, AI can propose unit tests, integration tests, and edge cases that developers might overlook.
Yet, as Tom emphasized, there are risks when AI is treated as the pilot:
The bottom line: AI augments engineering but doesn’t absolve responsibility. The core takeaway from our chat was simple but profound: You must remain the pilot.
Software engineering is not just about writing code—it’s about making architectural decisions, balancing trade-offs, and ensuring ethical responsibility. AI can suggest, but it cannot own accountability.
As Tom put it, “AI can assist, suggest, and even refactor—but you remain the pilot.” That distinction is vital. Whether you’re using GitHub Copilot, prompting LLMs for design patterns, or architecting AI-native systems, the human engineer must remain in control.
To avoid issues like security vulnerabilities, subtle logic bugs, or bloated codebases, Tom and I discussed a few rules of engagement:
AI in software engineering is best understood as a copilot—a partner that enhances speed, accuracy, and creativity. But the pilot’s seat belongs to the human, who ensures that the journey is safe, ethical, and aligned with the mission. So the real challenge isn’t whether AI can code—it’s how we, as engineers, balance AI assistance with human accountability.
This article is based on an episode of AI Minute Mondays, where industry experts share insights on AI adoption, implementation, and impact across various domains. Watch the full conversation with Shish Shridhar above to dive deeper into the technical details and hear more about his journey in Retail and startups at Microsoft.
At oikyo.ai, we help organizations navigate the complexities of AI adoption, from strategy and platform selection to implementation and compliance. Whether you’re in Retail, finance, or any other regulated industry, we can help you build AI systems that are not just powerful, but trustworthy.
Contact us to learn how we can support your AI journey, or explore our services to see how we help organizations accelerate trusted AI adoption.
Interested in more AI insights? Subscribe to our newsletter or follow us on LinkedIn for the latest in responsible AI adoption.
]]>